i VALUE CREATION IN PROXIMITY TO U.S. LIGHT RAIL TRANSIT STATIONS William Leslie Bishop A dissertation submitted to the faculty at the University of North Carolina at Chapel Hill in partial fulfillment of the requirements for the Doctor of Philosophy degree in the Department of City and Regional Planning. Chapel Hill 2018 Approved by: Roberto G. Quercia Nikhil Kaza Noreen McDonald Daniel Rodríguez Yan Song
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
i
VALUE CREATION IN PROXIMITY TO U.S. LIGHT RAIL TRANSIT STATIONS
William Leslie Bishop
A dissertation submitted to the faculty at the University of North Carolina at Chapel Hill in partial fulfillment of the requirements for the Doctor of Philosophy degree
Tan, Janssen-Jansen, & Bertolini, 2014). Such a virtuous cycle is illustrated in Figure 11.
Investment in (capacity and accessibility creating) transit infrastructure by transit agencies,
federal, state, and local government, and other public-sector actors can induce private sector
investment, development, and value creation.
Some part of any such value creation
may be captured by public sector
investors both to recoup investment
and to invest in additional transit
capacity. Subject to adequate
unsatisfied demand for transit,
increased capacity provides
improved accessibility and ridership
(as well as ancillary market demand
and activity) inducing additional value creation. Such value creation may include a variety of
urban amenities, such as those often embedded in transit-oriented development, that further spur
demand, and so on. Such virtuous cycles of investment, value creation, value capture,
reinvestment, and expansion of value-inducing capacity are possible because some part of the
value of enhanced mobility, accessibility, and other prospective amenities can become
capitalized into nearby land, commercial real estate, and housing prices (Agostini & Palmucci,
2008; Golub, Guhathakurta, & Sollapuram, 2012).
Notwithstanding the great potential for such value creation, and the potential for value
capture to offset public investment in transit infrastructure, results within the U.S. have been
1 Figures 1 through 3 are adapted from Guide to Value Capture Financing for Public Transportation Projects.
Figure 1: Virtuous cycles of value creation
5
mixed and often less than uniformly satisfactory both with respect to value creation and to value
capture. In response, the Transit Cooperative Research Program (TCRP), an applied, research
program that develops near-term, practical solutions to problems facing transit agencies, recently
sponsored development of the Guide to Value Capture Financing for Public Transportation
Projects recently published by the National Academies Press (2016).
Value capture is the public recovery of a portion of increased property value created as a result of public infrastructure investment. Common value capture mechanisms are impact fees, joint development, sale or leasing of air rights, land value taxation, station naming rights, negotiated exactions, parking fees, sales tax and special assessment districts (SADs), and tax increment financing (TIF). Given expanding demand for new transit infrastructure and scarce financial resources, U.S. transit agencies are increasingly looking toward innovative funding sources and strategies. Value capture is one of these innovative strategies (Page & Bishop, 2016).
Measuring transit-induced value creation
Capitalization effects, the extent to which the value of transit accessibility and/or
other transit proximity related amenities or benefits become capitalized into the market price of
real property, may be either positive or negative. Proximity to light rail transit stations may
increase property values because of enhanced mobility and accessibility as well as proximity to
other amenities. On the other hand, proximity to rail lines further removed from station area may
decrease property values because of nuisance effects and negative externalities such as noise,
vibration, and other environmental impacts (Armstrong & Rodriguez, 2006; Golub et al., 2012).
The literature on land and property values demonstrates a great deal of variability in the estimated change in values arising from rail investments… a meta-analysis on empirical estimates from 23 studies that analyzed the impact of rail on land/property value changes… show that a number of factors produce significant variations in the estimates. These include the type of land use, the type of rail service, the rail system life cycle maturity, the distance to stations, the geographical location, accessibility to roads, methodological characteristics, as well as whether the impacted area is land or property (Mohammad, Graham, Melo, & Anderson, 2013).
6
This study contributes to a growing body of literature regarding value creation
induced through proximity to light rail and/or other transit stations within transit-oriented
development (TOD) or within otherwise transit-influenced development (TID). Unlike hedonic
price models which estimate consumers’ marginal willingness to pay for amenities such as
proximity to transit, this study quantifies actual differential rates of aggregate annual value
creation within areas proximate to U.S. light rail transit stations compared to surrounding areas.
Additionally, this study contributes to an understanding of the impact on value creation of station
characteristics and transit agency implementation strategies. These include 1) early engagement
by transit agencies in strategic public-private partnership, 2) station location in terms of
development typology, 3) dominance of pedestrian or vehicular design, and 4) employment of
specific value capture strategies.
Transit-oriented and transit-influenced development, and value creation
Much of the literature relating to value creation and value capture has focused on
attributes of transit-oriented development in addition to the value-related effects of transit
capacity and accessibility. The viability and success of value capture strategy within TOD and
TID is explicitly dependent on the extent of differential value creation, as well as other
institutional, economic, market, and financial factors. While “it has long been recognized that
fixed transit infrastructure creates urban value in the property and land markets” (Cervero &
Kang, 2011; Rodríguez & Targa, 2004; Smith & Gihring, 2006), “there [have been] few
comprehensive assessment frameworks used to assess and capture the benefits…created”
(McIntosh, 2015). Any such value capture assessment framework is dependent on an
understanding of the extent of value creation potential, and the factors that affect value creation.
7
Extensive research has been conducted regarding institutional and inter-institutional
factors such as the need for coordination between urban land use and transportation planning,
with particular emphasis on transit-
oriented development. The need for
such coordination may be particularly
acute in the context of TOD, transit-
influenced development, and value
capture opportunity where many
actors must cooperate to realize
optimal outcomes.
Figure 2 illustrates the idea that once new direct investment in transit infrastructure is
effected through a transit agency (or equivalent) private sector developers (and investors and
speculators) may respond in a manner that creates or otherwise results in value creation
surrounding that infrastructure. In many cases, some portion of that value creation can be
construed as value premium in the sense that land or other real estate assets command higher
prices (value) than would be the case in the absence of the infrastructure investment. Any such
value premium creates opportunity for the public-sector investor(s) to capture some part of that
value to provide a return on or a partial return of the public investment. So long as the extent of
value capture does not exceed the infrastructure induced premium, such a revenue source
(exaction) should not create a competitive disadvantage in the market.
Value capture might be used to return revenue to other public entities such as local
government which may have invested (directly or indirectly) in transit supportive municipal
infrastructure. Value capture benefits (revenue) realized by local government may be used both
to invest in other public policy objectives (such as affordable or workforce housing for example)
Figure 2: Investment, value creation, value capture and transfer
8
and/or to provide additional value creating incentives to developers (infrastructure related impact
fee credits or offsets for example).
Effective value capture strategy can buttress a virtuous cycle of value creation
(Huxley, 2009; Levinson & Istrate, 2011). The potential for and extent of value creation within
TOD may be dependent, in part, on the extent of cooperation and strategic engagement between
transit agencies, local government, other public-sector agencies and interests, private landowners
and developers, and other private sector interests, as depicted in Figure 21. Transit agencies
seeking to benefit from value capture following direct investment in new transit infrastructure
must rely on the cooperation and engagement of private developers and local government in
providing additional investment and negotiating market appropriate value-optimizing
entitlements. Developers and local governments may also cooperatively engage with other
public-sector actors such as housing authorities to participate in additional value-creating
investment. The engagement referred to here is not merely that which is often undertaken in the
interest of balancing multiple interests toward the end of building consensus or acceptance
(Kaza, 2006), but that which is required to align strategic interests in long-term value creation,
particularly in the context of unknowable future economic and market conditions (Zapata &
Kaza, 2011). The extent and complexity of cooperative and strategic engagement required of
multiple public and private actors, each constrained by market forces and requirements of public
or private finance, is illustrated in Figure 31.
Experience suggests that frequent institutional reluctance to engage in strategic
partnership toward mutually beneficial TOD/TID value creation may result from cultural,
1 Figures 1 through 3 are adapted from Guide to Value Capture Financing for Public Transportation Projects.
9
institutional, administrative, and legislative forces as well as divergent market and non-market
incentives (Kaza, 2013). The institutions and parties that must cooperate in the interest of
maximizing value creation often operate from within largely isolated silos of language,
perspective, and vision.
Disinclination or reluctance to engage, cooperate, and plan and act together may
result from failure to recognize mutual benefits, and from very different understanding of the
meaning of plans themselves (Kaza, 2008). Overcoming such reluctance may require a more
thorough understanding of the nature and scale of potential benefit to all parties resulting from
both inter-institutional engagement, and in realization of significant incremental value creation
through joint planning and joint or coordinated action. Transit agencies and other public-sector
actors may benefit from a more thorough understanding of the tools and requirements necessary
to achieve such value creation (Kaza & Hopkins, 2007). Transit agencies may forgo beneficial
strategic partnership and market
engagement in favor of laissez-faire
reliance on the presumption that
market response to new or enhanced
transit infrastructure investment and
services will result in desired value
creation. Real estate price premiums
sometimes associated with TOD/TID
are frequently generalized and
interpreted as proxy for value creation
in the aggregate. Figure 3 illustrates the sort of strategic engagement and/or partnership required
to maximize private-sector value creation in response to public-sector infrastructure investment.
Figure 3: Inter-entity engagement and strategic partnership
10
Transit agencies, other public-sector agencies, and private-sector both invest directly in value
creation and stand to benefit (participate) in such value creation either directly or through value
capture. Strategic engagement, if not formal partnership, between these entities is required to
maximize the value creation opportunity. The nature of such partnership or engagement is
constrained by the requirements and peculiarities of the (dynamic) public or private finance
environments in which they operate. Land owners, speculators, and investors, and other public-
sector entities may benefit from value creation even if they do not invest in it. Notwithstanding
this, significant benefit may result from strategic engagement between primary public and private
sector actors and other stakeholders. Thorough understanding of the causes, requirements, nature,
and extent of transit infrastructure-induced value creation is both currently inadequate and
desirable.
Quantifying effects of proximity to light rail transit stations on assessed
valuation over time
This analysis seeks to test the hypothesis that proximity to light rail transit stations
resulting from public infrastructure investment results in higher rates of development and aggregate
value creation than occurs over the same periods of time in locations further removed from transit
stations. We have long observed that facilities providing for transportation of people and goods to
and from fixed geographic locations induce investment in transportation-supportive and other
infrastructure and result in concentrations of commercial activity. This has been true of harbors,
seaports, the mouths and confluences of rivers and navigable waterways, rail junctions and termini,
as well as Interstate and other highway junctions and interchanges. Similarly, research has identified
value creation effects associated with bicycle and pedestrian trail facilities as with conventional
The tendency for people to congregate, and for commerce and other human activity to
become concentrated, in centers or nodes of high activity where transportation has been facilitated
(and where direct transportation - or commuting - costs have been minimized), and for land rents to
increase with density and activity, is consistent with theories of urban spatial structure (Alonso, 1964;
Muth, 1979). Both economic theory and many hedonic price models extant in the literature suggest
that consumers who would benefit from living and/or working, shopping, or recreating in proximity
to transit stations should be willing to pay for such proximity. Such willingness to pay for
accessibility should create economic opportunity for those who would provide (develop) residential,
office, retail, and other real property improvements in proximity to transit, when and where the value
of those (consumer preferential) price premiums exceed the cost(s) associated with supplying the
amenities and specific bundles of goods demanded by consumers.
Although assessed valuation is a somewhat sluggish and imperfect measure of underlying
market value in real time, it is a useful and important measure of value for purposes of implementing
many value capture strategies and for considering related public policy objectives. Methodologies
associated with value assessment for ad valorem tax purposes present several concerns (particularly
with respect to assessed valuation as a real-time proxy for market value). These concerns are
addressed under “Limitations” below. Assessed valuation reduces or eliminates other concerns,
however, such as sample selection biases within price indices (Jud & Winkler, 1999), and provides a
number of practical advantages.
One attractive feature of the assessed-value method is the ability to efficiently incorporate property- and location-specific information from potentially every single … property that exists in a location as a data point. In addition, we appreciate that an expert assessor may be able to capture value adjustments not typically measured by the set of explanatory variables used in a standard hedonic type of estimation…In addition, the comprehensive nature of the database allows us to segment the data by value or geographical region and compare the price changes (Gatzlaff & Holmes, 2013).
12
This analysis quantifies differential annual rates of change in assessed value and folio
density between areas of treatment (immediately proximate to stations) and surrounding areas
(control) for 229 light rail transit stations along 21 LRT lines in 14 U.S. transit systems between 2005
and 2015, and evaluates a number of station-specific, transit agency, and demographic characteristics
in an effort to explain variation in differential value creation. Note that several of the subject transit
systems include stations on more than one LRT line.
Effect of station-specific characteristics and transit agency initiatives on
variation in differential rates of light rail transit influenced value creation
Differential rates of value creation within transit (station) areas of influence vary
significantly both between transit lines and markets as well as within the same transit line. This
analysis seeks to identify the extent to which specific station characteristics are associated with
variation in differential rates of value creation. Specific station characteristics include station
design (e.g. elevated, at grade, open cut, underground, etc.), dominant station character (e.g.
walk-and-ride or park-and-ride), the number of parking spaces provided at each station, and a
range of station locational typologies (e.g. downtown – Central Business District (CBD), urban
center, urban neighborhood, suburban town center, suburban neighborhood, campus,
entertainment, special).
Value creation (and any potential for subsequent value capture) is influenced by other
factors as well. Robust value creation is dependent on a number of requisites including:
real estate market vitality, accommodative zoning and land use entitlements; and development of project- and context-specific financial strategies that are feasible and incentivize and reinforce value creation; and institutional capacity on the part of transit agencies, local governments, developers, and other partners working together to maximize value creation and value capture” as represented in Figure 4 (Page & Bishop, 2016).
13
Economic conditions fluctuate significantly over time. Real estate market conditions vary
from time to time and place to place. Market conditions specific to each metropolitan market are
captured within Transit System designation. Responses to transit agency survey questions (described
below) address 1) the extent of public-private engagement [represented as Public-Private
Enterprise arrow in Fig. 4], 2) institutional capacity, and 3) regulation regarding value capture to
some degree. Although institutional capacity and regulation affect value capture potential more
directly that value creation, they may inform the extent to which transit agencies are focused on
maximizing value creation early and strategically.
Figure 4: Conditions precedent to value creation and capture, Adapted from Guide to Value Capture Financing for Public Transportation Projects
This analysis seeks to identify the extent to which specific transit agency objectives
and initiatives are associated with variation in station-specific rates of differential value creation over
time. Surveys were solicited from senior planners and managers from the 14 transit agencies
represented in this study. Survey respondents were asked to classify stations within standardized
design, locational, and functional typologies. In addition, respondents were asked to provide
background on transit line project and planning objectives, initiatives related to value creation and
14
value capture, strategic engagement with entities outside the transit agency, and the relative
importance of various initiatives to achievement of transit (i.e. planning and investment) objectives.
Brief summary of findings
Differential rates of assessed value creation varied widely across transit systems and
individual stations. Where treatment areas were defined such that there was no overlapping of
adjoining treatment areas, Transit Areas of Influence (TAIs) in the top 20% of differential value
creation experienced average annual growth rates 30.67% faster than that of control, whereas
TAIs in the bottom quintile experienced negative average annual differential growth rates
(1.25% less than those within control areas). Significantly positive differential value creation
was concentrated within a small number of transit systems and within a relatively few number of
stations along several of the lines studied.
Differential rates of assessed value creation are found to accrue disproportionately to
improvements (and to folio density) rather than to land. In the aggregate, assessed
improvement(s) values grew 4.35% faster in treatment areas than in control areas, whereas
assessed land values grew only 3.28% faster within treatment areas. The extent to which
treatment explained variation in differential value creation was roughly twice that for assessed
improvement(s) value and that for assessed land value (approximately 2% compared to
approximately 1%). Additionally, although the treatment-time interaction effects were significant
over the entire period for assessed improvement values, the effect was not significant in any
individual year for assessed land value.
Significant covariates included transit system (location), accounting for 19% of the
differential rate of change in total assessed value per acre over time; per capita income (at the
15
census block-level), accounting for an additional 3% of variation; station design, accounting
2% of variation; transit agency perceptions of public-private value creation strategies as
important 1) to the success of transit infrastructure investments and projects, and 2) to the
success of transit-influenced or transit-oriented development, accounted for an additional 1% of
variation each. Significantly positive differential value creation occurred predominantly near
at-grade stations.
Of numerous demographic covariates evaluated, only per capita income and vehicles
per household were significant, predicting 4% and 2%, respectively, of the variation in
differential rates of change over time (the interaction of covariate and time) between treatment and
control groups.
Where treatment areas were defined such overlapping of adjoining treatment areas
was allowed, differential rates of assessed value creation per acre were similarly highly varied
across transit systems. Differences between compounded annual rates of value creation within
treatment and control areas ranged from -1.16% to 7.82% (3.21% average across all stations,
regardless of system).
Contribution to literature and practice
This study attempts to bridge some part of the gap between the generally accepted
understanding of potential for transit price/value premium (i.e. some consumers willing to pay
price premium to live/work in proximity to transit-oriented development, under certain
circumstances), and a larger understanding of differential light rail transit induced value creation
in the aggregate. This distinction may be important and useful to policy makers deciding when
and where to make infrastructure investments or how to maximize the realizable value of those
investments; transit professionals seeking to lay the groundwork for optimal value creation
and/or value capture; and financiers undertaking to finance such investments. Many transit
16
professionals, and other advocates of public transit in general and/or LRT, may be enticed by the
notion that investment in new light rail transit capacity and stations can result in transit value
premiums capitalized into real estate value. This premium (consumers’ willingness to pay,
realized in some locations during some periods under certain circumstances) may become
conflated with the aggregate value creation on which policy makers and financiers must rely in
order to realize value capture as a viable source of infrastructure finance. The fact that specific
markets or sub-markets may respond to new LRT stations with significant apparent price
premiums may become misinterpreted as an indication that such market responses occur
spontaneously and/or more or less uniformly. This may devolve into a sense and expectation that
“if we build it they will come,” setting up disappointment when value creation fails to materialize
uniformly or robustly.
This study spans those years that were significantly and adversely impacted by the
Great Recession. It appears that market forces and characteristics other than those captured
within the covariates in this study influence differential value creation in proximity to new light
rail transit systems to a significantly greater extent than the treatment (i.e. new
infrastructure/investment) itself. This underscores the importance of understanding and
underwriting such market condition and factors before undertaking projections with respect to
anticipated levels of value creation.
This study is intended to inform public policy and professional practice in the U.S.
with respect to planning for and realizing value creation, particularly in the context of
prospective investment in new LRT infrastructure. Policy considerations may include not only
the question of whether to invest in new transit infrastructure along a particular alignment in a
particular place at a particular time, but what transit technology is optimal. The question of
transit mode/technology involves not only its effectiveness and utility as transportation, but its
17
capital and operation costs both before and after consideration of potential value creation and
value capture, and its potential to induce economic development. Light rail transit may not be the
low-cost transportation solution in all settings. Justification of the significant capital investment
required of LRT may require reliance on significant value creation and related fiscal impact and
economic benefits. A more thorough and context-specific understanding of the nature and causes
of light rail transit-induced value creation is desirable.
18
CHAPTER 2: LITERATURE REVIEW
That growth in real property value in which policy makers and financiers of public
transit infrastructure are interested occurs over long periods of time and must be stable and
enduring. Its conditions precedent must be well understood. Policy makers, planners, and
financial advisories and underwriters must be able to anticipate with some confidence when,
where, and to what extent value will be created in response to new infrastructure investment.
Dominant themes in the literature suggest that although uneven, consumers are
generally willing to pay some price premium for real estate proximate to light rail or other transit
stations (including Bus Rapid Transit, or BRT) and/or within transit-oriented or transit-
influenced development. Some studies have suggested that consumers’ revealed willingness to
pay a price premium may be related as much to a specific range of urban and lifestyle amenities
as to transit accessibility; whether or not those amenities are located near transit stations. Most of
these studies have employed hedonic price models to disaggregate consumer preferences for
various individual characteristics of highly heterogeneous properties; which characteristics are
not traded individually in the marketplace. Some discussion of hedonic models follows within
this literature review, as does review of literature addressing appropriateness of assessed
valuation as a measure of value.
Many studies have been undertaken to estimate the real property value effects of
proximity to light rail (and other) transit stations. (NEORail, 2001) There have also been
extensive reviews of this literature (Cervero & Aschauer, 1998; Cervero, Ferrell, & Murphy,
Mojica, 2009); land value impacts of BRT (Rodriguez & Mojica, 2008); planning for development
in accommodation of BRT (Gakenheimer, Rodríguez, & Vergel, 2011); the relationship between
urban form and station boardings (Estupiñán & Rodriguez, 2008); examination of the reciprocal
relationship between BRT and the built environment in Latin America (Vergel-Tovar, 2016); and
public transport investments and urban economic development (Heres, Jack, & Salon, 2014).
Other BRT studies examining development patterns (Cervero & Landis, 1997; Fogarty
& Austin, 2011); inducement of TOD, or as prospectively cost-effective alternatives to LRT
include: planning for BRT as a modal alternative to “Light Rail Lite” (Hoffman, 2008); leveraging
TOD with BRT investment (Cervero & Dai, 2014); BRT and urban development (Rodriguez &
Vergel, 2013); BRT as a substitute for LRT (Sislak, 2000); comparison of BRT and LRT fixed
guideway systems (Biehler, 1989); a review of BRT literature (Deng & Nelson, 2011); real estate
impacts from fixed rail and BRT (Kannan, 2011); and impact of bus transit centers on values of
nearby single-family residential land in Houston, Texas (Lewis & Goodwin, 2012).
Light rail transit and transit-oriented development (LRT and TOD)
Transit-oriented development (TOD) is one specific type of the many potential forms of
transit-influenced development. TOD is typically composed of vibrant mixed-use development that is
29
amenity-rich and features proximity to transit. Many multimodal features are included in TOD,
including pedestrian and bicycle improvements. Numerous studies have demonstrated that under
certain circumstances, TOD can command higher sales prices and rents for a variety of property
types. The opportunity for value creation and subsequent value capture will vary by transportation
network and station characteristics. Unique characteristics of each transit line and station area will
influence the potential for value creation and capture (Song, 2002).
This study addresses value creation in proximity to LRT stations without explicit
distinction between TOD, other forms of transit-influenced development, or station areas in which
there has been little discernable transit-induced value creation at all. TOD, the value it can create, and
the price premiums it can command, have garnered a great deal of attention and inspired a great deal
of academic and commercial study. Although this study does not focus on TOD per se, the subject
commands some acknowledgment. In many cases, some significant part of that value capitalized into
real estate prices in proximity to transit stations may derive as much or more from TOD or TOD-like
urban amenities as from transit accessibility itself (Song & Knaap, 2003).
TOD involves:
creating attractive, memorable, human-scale environs with an accent on quality-of-life and civic spaces. Increasingly, projects built around up-and-coming transit nodes, like Dallas’s Mockingbird Station, Portland’s Pearl District, and Metropolitan Chicago’s Arlington Heights, are targeted at individuals, households, and businesses seeking locations that are vibrant and interesting; these places usually have an assortment of restaurants, entertainment venues, art shops, cultural offerings, public plazas, and civic spaces” (Cervero, 2004).
The many lifestyle and urban amenity benefits that may be realized from within transit-
influenced projects … result not only from transit access but also from particularly complex and
compact mixed-use real estate development and occupancy. The complexity and intensity of TOD
projects can create risk and discourage value-maximizing real estate development and private-sector
30
investment. TOD often requires significant up-front investment in infrastructure and common
amenities (Carlton, 2009). Many of the requirements for successful value creation within TOD fall
outside the control of developers and require engagement, collaboration, and partnership with transit
agencies and local governments (Hale, 2008; Hale & Charles, 2007; Hale, 2013). A great deal of
cooperative engagement and strategic partnership is required in both planning and execution. A
paradigm shift is needed “from current practice of small scope—ad hoc, technical solution driven—
planning approach towards a new practice that considers a broad network scope—strategy driven—
planning approach” (p. 1, Arts, Hanekamp, and Dijkstra, 2014). These considerations underpin the
survey questions posed to transit agency officials in the present study regarding strategic engagement,
value creation strategies, and transit agency goals and objectives.
Numerous studies have estimated the impact of TOD and/or specific elements of TOD on
various classes of real property values. Examples include measuring the impact of suburban TODs on
single-family home values (Mathur & Ferrell, 2013); development density (Litman, 2014); economic
development impacts (Litman, 2010); and effects of pedestrian elements of TOD (Bartholomew &
Ewing, 2011). In general, these studies have identified positive value creation or increased consumer
marginal willingness to pay (price effects) related to TOD and many of its common attributes or
constituent parts (Clifton, Ewing, Knaap, & Song, 2008).
Quantifying effects of proximity to light rail transit stations on assessed
valuation
Many studies have utilized indicators for price/value other than assessed valuation. These
have included residential or office rental rates, realtor listing prices, deed records, proprietary records
of sample transactional data, repeat sales records, and market price indices, in addition to assessed
value. Several papers have supported the validity of assessed valuation data as a measure of value and
for purposes of developing price indices (Case & Wachter, 2005; Clapp & Giaccotto, 1992; Gatzlaff
31
& Holmes, 2013; Jud & Winkler, 1999).
Although assessed valuation data can suffer from several drawbacks as proxy for market
value, they have advantage that they are available for all revenue acreage within subject treatment
and control areas and are subject to consistent methodologies within, if not between, jurisdictions in
this study. Assessed values are perhaps “the most comprehensive and reliable government data
source for residential property values” (Hess & Almeida, 2007). Another significant advantage of
assessed valuation data is that they incorporate the entire universe of (taxable) properties whereas
other datasets based on real estate transactions reflect only those properties that have been traded in
the marketplace.
Examples of studies that have employed assessed valuation data for similar purposes
include: The Impact of the Miami Metrorail on the Value of Residences near Station Locations, 1990
Dade County property tax records (Gatzlaff & Smith, 1993) and Impact of Proximity to Light Rail
Rapid Transit on Station-area Property Values in Buffalo, New York (Hess & Almeida, 2007). Other
studies employing assessed valuation to estimate the effect of transit station proximity on property
values include the following examples by market: Atlanta, Georgia (MARTA), DeKalb County tax
assessor (Nelson, 1992); Dallas, Texas, (DART), Dallas County Central Appraisal District
(Weinstein et al., 2002); Miami, Florida (Miami Metro Rail), Dade County property tax records
Sphericity was examined using Mauchly’s test. None of the variables exhibited
sphericity (p < .001 in all cases), thus violating a requisite ANOVA assumption. To compensate,
the Greenhouse-Geisser correction, which adjusts the test statistic for violations of sphericity,
was applied to interpret all analytical results.
Homogeneity of variance and covariance matrices were assessed next. Homogeneity
of variance was assessed using Levene’s test, and was met for some variables (p > .05), but not
met for others (p < .05). Homogeneity of covariance was assessed using Box’s M test. Box’s M
test could not be calculated due to the large number of variables in the model. Because
homogeneity of covariance could not be assessed, the more conservative Wilk’s lambda
coefficient was interpreted to compensate for any possible violation of the homogeneity
assumptions. Wilk’s lambda is a more conservative statistic in that it is less likely to indicate
significance, but is also less susceptible to the possibility of Type I error (i.e. the incorrect
rejection of a true null hypothesis) introduced by the violations of homogeneity (Stevens, 2016).
Treatment (i.e. an indicator variable denoting treatment or control group) was the
principal independent variable in each of the five models. Covariates included a transit system
(location) identifier and a variety of demographic characteristics (selected as described above)
including: median household income, per capita income, gross rent burden greater than 30%,
persons per household, percent dwellings vacant, percent dwellings owner occupied, total
43
vehicles per household, total vehicles per owner-occupied household, and total vehicles per
renter-occupied household. Except for the first model, covariates included those identifying
station-specific characteristics such as station typology, character, design, and number of transit
agency provided parking spaces. [Station design was dichotomized due to small group
frequencies, resulting in a variable with two levels consisting of at grade and other]. Additional
covariates addressed transit agency design and development objectives (e.g. reducing roadway
congestion), employment of value creation strategies (e.g. land use, zoning, and entitlement
enticements), and employment of specific value capture strategies.
Tables 1a and 1b identify variables of interest as well as covariates—including those
identifying station location, station-specific characteristics, transit agency perspectives and
initiatives, and as demographic characteristics at the census block-level.
44
Table 1: Variables, measures, and sources
45
Table 1b: Variables, measures, and sources continued
Geographic Information Systems
Locations of the fourteen (14) U.S. LRT systems are considered in this analysis.
Subject transit lines include Charlotte, CATS Blue Line; Dallas, DART Green Line; Denver,
RTD Blue, Orange, and Purple Lines; Houston, MetroRail; New Jersey Transit, Hudson-Bergen
and River Lines; Los Angeles Metro Expo and Gold Lines; Minneapolis, Metro Blue Line;
46
Norfolk, Tide Light Rail; Phoenix, Metro Light Rail; Portland, TriMet MAX Green, Red, and
Yellow Lines; San Diego, Oceanside Sprinter; Seattle, Link Light Rail; and Salt Lake City,
TRAX Red Line, Redd Line-Airport, and Green Line. These are depicted in Figure 5 below.2
Figure 5: 14 U.S. transit systems comprising the subjects of this analysis
Treatment Areas
Selection of treatment areas surrounding the 229-subject LRT station areas was based
on the primary transit ridership catchment areas within TAIs as defined by the American Public
2 The St. Louis MetroLink light rail system, with lines and station in both Illinois and Missouri was dropped from the
analysis due to extensive missing data and other data problems.
47
Transportation Association (APTA). APTA defines TAIs as those “spatial areas in which transit
stops and stations typically have the greatest impact on land use and development and from
which there is high potential to generate transit ridership” APTA further delineates TAIs “for
purposes of influencing decisions about private and public investments and services.” Primary
catchment areas, defined as those contained within ½-mile radii from station centers (and
including the “transit core”), are those:
within which land use and urban design features, as well as the ease and directness of access to the stop or station, have a substantial impact on transit ridership and pedestrian access. The primary catchment area may generate a significant portion of total transit trips to and from the stop or station. (APTA, 2009)
Figure 6: Defining Transit Areas of Influence APTA SUDS-UD-RP-001-09 (APTA, 2009)
Primary Catchment Area – Light Rail Transit: ½-mile Radius
Coordinates for latitude and longitude were collected for the centroids of each of the
229 subject stations. ArcGIS was employed to define individual treatment area geographies
48
within a ½-mile radius of each station centroid. Within the ANOVA part of the analysis, and in
cases where treatment areas overlap because station centroids are within 1-mile of each other,
overlapping treatment areas were cropped employing a Thiessen polygon and GIS union method
to eliminate overlapping areas and to avoid double counting of folio values within treatment
areas. The concern here was that double (or triple) counting values within treatment areas (in the
sense that the same folio could appear within ½ mile of two or more station centroids) might
positively bias apparent rates of differential value creation.
Thiessen polygons were constructed, then a union function was run with the 1/2-mile
buffers to create non-overlapping territories for each transit stop. Thiessen polygons are those
with boundaries defining areas closest to each point relative to all other points. This method for
defining treatment areas where there would otherwise have been overlap was used for three
reasons: 1) GIS processing complexities required to bisect areas of overlap precisely within ½
mile transit stop buffers, 2) time intensiveness otherwise required to manually interpret and
delineate boundaries, and 3) ease of replicating results for extended future research or study. A
detailed description of this and other GIS procedures has been provided by Phillip McDaniel,
GIS Librarian, and appears as Appendix III.
Figure 7 provides an illustrative example of discrete treatment area definition
employing the Thiessen polygon and Union method.
49
Figure 7: Treatment areas defined exclusive overlaps using Thiessen polygon and union method
Treatment areas not required to be cropped to eliminate overlapping with contiguous
station areas comprise 502.66 acres within the ½-mile radius. The station area with the smallest
geographic area, due to elimination of overlaps, comprises 137.60 acres (the Nicollet Mall station
on the Blue Line in the Minneapolis Metro system). In total, the 229 treatment areas comprise
93,847.76 acres, or a mean of 409.82 acres per station area. Individual folio records are selected
into treatment for all folios with a geographic centroid falling within the treatment area. All
others are excluded.
50
Control Areas: Geographies surrounding subject TAIs
Control areas surrounding the 229-subject LRT station areas are defined as those
within and bounded by 1-mile and 2-mile radii from the centroids of each of the 229-subject
stations. Unlike treatment areas, no cropping of control areas is undertaken to eliminate
overlapping control geographies. The size of specific control areas varies not because of
contiguous control geographies, but because of the elimination of nearby treatment and other
buffer areas from control.
To minimize the extent of treatment confounding control, and to minimize the extent
to which particular folios (large land parcels in particular) might overlap treatment and control
areas, treatment and control areas are separated by a ½-mile buffer as illustrated in Figures 9 and
10. In addition to the buffers surrounding treatment areas, a ½-mile buffer from the centerline of
subject LRT lines is excluded from control areas. Rationale for elimination of folios within the
½-mile buffer area surrounding stations is that (notwithstanding APTA definition of primary
catchment areas within TAIs), any treatment effect of proximity to transit stations including the
benefits of mobility and accessibility being capitalized into real estate and reflected through
market prices and assessed valuation, does not and would not end abruptly at the ½-mile radius.
One practical effect of this, in the event that differential value creation effects are positive, is that
comparing rates of change between control areas (at greater distance from stations) and those
areas most impacted/benefited by proximity to stations is likely to overstate any aggregate
treatment effect. On the other hand, this study is likely to underestimate aggregate differential
value creation insofar as there is a positive (value-creating) treatment effect, it would likely
extend beyond the arbitrary ½-mile radius. Conversely, where differential value creation effects
of proximity to LRT stations may be negative, the effects would be reversed.
51
Rationale for elimination of folios within the ½-mile buffer along rail line corridors
outside of TAIs is that proximity to LRT lines and rail traffic may have a negative effect on the
value of property not proximate to stations (within treatment areas) (Armstrong & Rodriguez,
2006). Folios within these buffer areas are eliminated to avoid overstatement of any
prospectively positive treatment affect. The following two figures illustrate an outlying control
area (in two large parcels) surrounding a central treatment area (the circle at center) and other
nearby treatment areas not associated with the identified control area. The distance between the
treatment area(s) and the control area comprise exclusion areas.
Figure 8: Representative treatment and corresponding control areas, exclusive of folio centroids.
52
Figure 9: Representative treatment and corresponding control areas, inclusive of folio centroids.
Control areas range in size from 1,937.73 acres (the Garfield Avenue station on the
Hudson-Bergen line in the New Jersey Transit system) to 6,031.87 acres (the Bordentown and
Florence stations on the River Line in the New Jersey Transit system, and the Tukwila
International Blvd. station on Seattle’s Link Light Rail system). In total (including overlapping
control geographies), the 229 control areas comprise 826,526.07 acres, or a mean of 3,609.28
acres per station area. Individual folio records are selected into control for all folios with a
geographic centroid falling within the control area. All others are excluded.
Quantifying effects of proximity to light rail transit stations on assessed
valuation
This study employs a difference-in-differences approach to estimate differential rates
of value creation within ½-mile of LRT transit stations (treatment), compared to those within
areas between 1-mile and 2-miles from station centroids (control) over the 11-year period from
2005 through 2015. Year over year differences in assessed values within designated treatment
areas are estimated on a percentage change basis and compared to those within control areas.
53
Differences between treatment and control areas are estimated for each of four values aggregated
from the 2,080,619 individual (annual) folio records within treatment areas and 15,507,734
records within control areas. The four observed values include those for the assessed value of
land (T_AVL for treatment and C_AVL for control), assessed value of improvements (T_AVI
and C_AVI), total assessed value (T_AVT and C_AVT), and folio count (T_COUNT and
C_COUNT). All values were converted to densities per acre based on the acreage of individual
treatment and control areas.
Difference-in-differences is a quasi-experimental design that compares observed
outcomes after treatment (commencement of transit service), and in some cases before treatment,
within defined treatment areas (1/2-mile radius Primary Catchment Areas within TAIs), with
control cases comprised of folios located at greater distance from stations. The DD design seeks
to isolate the effect of treatment (the difference of interest) from any difference that would have
resulted in the absence of treatment (Meyer, 1995). Kuminoff, Parmeter, and Pope have found
“the difference-and-difference model with spatial dummies [to be] most accurate in identifying
[marginal willingness to pay] in situations where there is a time dimension to the data…”
(Kuminoff, Parmeter, & Pope, 2009).
The DD design can help address omitted variable bias, long a concern with respect to
hedonic price models. We may expect (even desire) transit amenities to be correlated with
unobserved neighborhood characteristics (or those which are difficult to capture with quantitative
data). We can “use spatial fixed effects to help absorb the confounding influence of omitted
variables and … quasi-experimental methods [such as DD] to purge time-constant omitted
variables from [models]” (Kuminoff, Parmeter, & Pope, 2010).
54
Mixed model ANOVA
This study employs multiple mixed ANOVA models (one-within and one-between
with covariates), where the within-subject factor is time (11 time periods for each dependent
variable), and the between-subject factor is group (treatment or control) as categorical
independent variable. The dependent variable in the principal model consisted of total assessed
value per acre per year from 2005 through 2015. Additional models with dependent variables
consisting of assessed value-land per acre, assessed value-improvements per acre, and folio
density per acre are addressed in greater detail in the “Supplemental Studies” section below.
Assessed value and folio densities are expressed in per acre terms to facilitate level comparison
between treatment areas, and treatment and control areas, of significantly different sizes. The
independent variable was treatment group (treatment or control), and covariates included those
controlling for location, station and transit agency specific characteristics and various
demographic characteristics.
The effect of group (treatment or control), and the interaction effect between time and
group is used to test the hypothesis that assessed value of real property in proximity to light rail
transit stations will increase faster over time than surrounding properties located at greater
distance from stations. The interaction constitutes a difference-in-differences analysis
(differences over time between differences between treatment and control).
The first mixed ANOVA model estimated the significance and effect size on values of
the interaction between treatment and time controlling for location (by transit system), capturing
local economic and market conditions, and demographic characteristics derived from 2010
American Community Survey data at the census block-level.
A statistically significant group effect indicates that there are significant differences
between treatment and control within common time periods. Although calculation of such
55
“main” effects is necessary within the mixed ANOVA model, these have little practical
interpretive value in the context of this study.
The mixed ANOVA analysis can determine whether there is an overall between-
subjects effect of treatment (proximity to station or treatment), whether there is an overall within-
subjects effect of time (change in assessed value per acre from year to year during the period
2005 through 2015, regardless of proximity to station), and whether there is significant
interaction between time and treatment. Evaluation of the time-treatment interaction will
determine whether there is a significant difference in rates of assessed value creation between
treatment and control areas over time. A statistically significant interaction between treatment
and time indicates that the values within the treatment group were significantly different than
those within the control group over time as a result of the treatment interaction.
Prior to analysis, the seven assumptions required for valid results from repeated
measures ANOVA were assessed and addressed ("Mixed ANOVA using SPSS Statistics," ;
Tabachnick & Fidell, 2007; Tabachnick et al., 2001). These include:
1. Dependent variable must be continuous; as is the case for all four measures of assessed
valuation.
2. Within-subjects factor or independent variable (time, in this case) should consist of at least
two categorical, related groups or matched pairs (treatment and control groups).
3. Between-subjects factor or independent variable (treatment and control) should each consist
of at least two categorical, independent groups.
4. No significant outliers in any group of your within-subjects factor or between-subjects
factor. (Obvious outliers were culled during data cleaning process.)
56
5. Dependent variable should be approximately normally distributed for each combination of
the groups of your two factors (within-subjects factor and between-subjects factor).
Normality was assessed through a Kolmogorov-Smirnov (KS) test. The KS test was
significant (i.e., p < .001 to p = .005) for each of the four dependent variables both for
treatment and control groups, indicating that normality cannot be assumed. However, when
sample sizes are large (i.e., N > 50), as in this dataset, the ANOVA is robust with respect to
violation of the normality assumption, so that analysis may be undertaken with valid results
(Howell, 2016; Pallant, 2013; Stevens, 2012; Tabachnick et al., 2001).
6. Homogeneity of variances for each combination of the groups of within-subjects factor and
between-subjects factor. Homogeneity of variance was assessed using Levene’s test, and
was met for some variables (p = .073 to p = .978), but not met for others (p = .010 to p =
.019). Homogeneity of covariance was assessed using Box’s M test. Box’s M test could not
be calculated due to the number of independent variables in the model. Given that
homogeneity of covariance could not be assessed, the more conservative coefficient
(Wilk’s lambda) was interpreted to compensate for any possible violation of the
homogeneity assumption, and to address the violations indicated by Levene’s test.
7. Sphericity, the variances of the differences between the related groups of the within-subject
factor for all groups of the between-subjects factor must be equal. Sphericity was examined
using Mauchly’s test. None of the variables exhibited sphericity (p < .001 in all cases). To
compensate, the Greenhouse-Geisser correction was used to interpret results of all analyses.
Time interaction effects
Interaction effects represent the combined effects of independent variables on
the dependent variable. In this case, we are most interested in differential value creation over
57
time. We are particularly interested, therefore, with the interaction of time (across the 11-years of
the study) with other independent variables such as station characteristics. When an interaction
effect is significant, the impact of one independent variable is dependent on another (time in this
case). One of the powerful utilities of ANOVA is its ability to estimate and test such interaction
effects.
Where we find statistically significant interaction effects, we cannot interpret
main (independent) effects without considering the interaction. In such cases, the effects of
independent variable are not themselves independent. A statistically significant interaction
between one independent variable and another (such as time) suggests that the effect of one
independent variable has been moderated or modified by the other.
Research questions and hypotheses
The primary question of this research is “Have higher rates of assessed value creation
over time resulted within proximity (i.e. primary catchment areas within transit areas of
influence) compared to similar locations at greater distance from new light rail transit
service and stations?”
The hypothesis anticipating this study was that higher rates of aggregate value
creation have resulted in proximity to transit stations, but that differential rates of value creation
would be inconsistent and highly varied across stations.
Total assessed valuation per folio, and the aggregate of total assessed valuation across
all folios, is the basis on which ad valorem taxes are levied and the most relevant metric
reflecting real property value for many fiscal impact, finance, and public policy considerations.
This analysis focuses principally on total assessed valuation for this reason. Additional questions
are considered with respect to other aspects of assessed valuation including:
58
§ Have higher rates of assessed land values resulted over time within proximity
(primary catchment areas within TAIs) than in similar locations at greater distance
from light new light rail transit service and stations?
§ Have higher rates of assessed improvement values resulted over time within
proximity (primary catchment areas within TAIs) than in similar locations at greater
distance from light new light rail transit service and stations?
§ How do differential rates of change in assessed values of land and improvements
differ from each other? This question is relevant, in part, because theory suggests
that mobility and accessibility benefits become capitalized into land values, not into
buildings and other improvements.
§ Have higher rates of folio (count/acre) creation over time resulted within proximity
(primary catchment areas within TAIs) than in similar locations at greater distance
from light new light rail transit service and stations?
Although we may conceptualize commencement of station operation (availability of
transit services) as the temporal point of treatment for purposes of distinguishing longitudinal
panel data pre-treatment and post-treatment, the effect of proximity to (existing or proposed)
LRT stations on real property values does not occur at a single, discrete moment in time. Land
and other real estate prices may reflect either positive or negative effects of such proximity years
before a system is delivered and a station is opened for operation and service. Real estate markets
respond to announcements of transit studies and alignments, determination of station locations,
commencement of engineering, funding commitments, and commencement of construction.
It does not necessarily follow that if land prices jump once a rail service begins that transit caused this appreciation. Spikes in land values could be attributable to other factors, like an upswing in the regional economy, improved highway conditions, or better schools. The challenge, then, is to control for such potential confounding factors so that the unique effects of transit proximity on land values can be partialed
59
out (Cervero & Duncan, 2001).
Panel Regression
Supplementing ANOVA, a second analysis was performed on the 200 of 229 station
areas for which there were no missing annual assessed valuation data. Regression analysis was
performed on panel data constructed based on a different research design premise, and definition
of treatment area, than that underpinning the ANOVA analysis reported above. Each of the
(502.66 acre) Transit Areas of Influence (treatment areas) within ½-mile of station area centroids
was considered in its entirety, regardless of whether or not that treatment area encroached into and
overlapped one or more adjoining Transit Areas of Influence. Whereas Thiessen polygons and a
GIS union method were previously employed to eliminate overlapping treatment areas and to
avoid double counting of folio values within treatment areas, no such “cropping” of treatment
areas was undertaken for this alternative analysis; this revised definition of treatment areas
includes every folio within ½-mile of station centroids.
60
Figure 10: Overlapping Transit Areas of Influence and Treatment Areas
Differential rates of total assessed value growth per acre, by station per year, were
regressed on a variety of prospectively explanatory variables to estimate the extent to which they
may cause differential rates of assessed value growth. Independent variables were comprised of
transit system, station characteristics (design, typology, character), transit agency goals and
objectives, value creation and value capture strategies, the relative importance to transit agencies
of various objectives and outcomes, and a variety of demographic characteristics, all previously
described. Given that many of the independent (prospectively explanatory) variables are time-
invariant with respect to stations, random-effects panel regression was employed (A Kohler & A
Kreuter, 2012). Practical experience, intuition, and results presented herein all suggest that
differences across stations, lines, and transit systems (metropolitan markets) unaccounted for in
these analyses, have influence on the rate of assessed value creation per acre.
61
CHAPTER 4: DATA
Previous studies evaluating assessed valuation data have relied on records from
individual tax assessors’ offices, or commercial data services such as Metroscan. Studies
evaluating representative sales or repeat sales data also rely on commercial data providers such
as TRW REDI, RLIS, Costar, First American Real Estate Solutions, and others. This study
employs assessed valuation data provided by CoreLogic. Headquartered in Irvine, California,
CoreLogic provides financial, real property, and consumer data and analytics to public sector,
private sector, and academic clients. CoreLogic has been accumulating annual assessed valuation
data within U.S. metropolitan jurisdictions for more than 30-years. As of the (2016) date of data
procurement for this analysis, 2015 was the “current” year of data availability. (Unfortunately)
CoreLogic maintains and makes available assessed valuation data only for the current years and
ten (10) preceding years. Data for prior years is disposed of. Assessed valuation data available at
the time of data acquisition was that for the period(s) 2005 through 2015. In addition to
discontinuation of data maintenance for periods more than 10-years prior to current year,
CoreLogic maintains the non-value elements of the assessed value folio records, such as building
characteristics, etc., for the current year only.
Based on the 11-year period of data availability, individual U.S. transit lines and
stations were selected for study based on relevant commencement of service dates. This selection
resulted in 286 light rail transit stations eligible for analysis. Elimination of all stations along the
St. Louis MetroLink line and various station along other lines for reasons of data availability and
appropriateness reduced the total number of stations subject to analysis to 229 (see “Limitations”
below, and a list of deleted stations in Appendix II).
62
Annual assessed valuation data by folio, geo-located by latitude and longitude per folio
centroid, was acquired from CoreLogic for years 2005 through 2015 for those FIPS codes
identified in Table 2.
Table 2: Transit systems and lines
Selection of all FIPS code relevant folio records into treatment areas yields a total of
154,829 folio records in 2005 increasing to 207,677 in 2015. Similarly, selection of FIPS code
folio data into control areas yields 1,198,661 folio records in 2005 increasing to 1,509,798
Met Area County ST FIPS System Line Opened
Charlotte Mecklenburg NC 37119 CATS Blue Line 2007
Dallas Dallas TX 48113 DART Green Line 2009
Denver Denver CO 08031 Denver RTD C Line - Orange Line 2002
Arapahoe CO 08005 E Line - Purple Line 2006
Douglas CO 08035 F Line - Red Line 2006
H Line - Blue Line 2006
Houston Harris TX 48201 METRO Rail METRO Rail 2004
Jersey City Hudson NJ 34017 Hudson–Bergen Light Rail 2003
Los Angeles Los Angeles CA 06037 Metro Rail light rail Gold Line To Pasadena 2003
Gold Line Eastside Ext 2009
Minneapolis Hennepin MN 27053 METRO Blue Line 2004
experienced an average increase of 306.72% over the period. Whereas stations within the 20% of
transit systems exhibiting the lowest rates of increase in total assessed total values per acre within
treatment areas over time experienced an average decrease of 12.46%.
Although differential value creation rates of stations within the second and third
quintile were also positive, these rates were very highly skewed. High rates of differential value
creation occurred most markedly only in the first quintile. Differential rates of total assessed
value creation in the other 80% of stations could fairly be described as “flat,” as illustrated in
Figure 11.
Figure 12: Differential (mean) rates of change in total assessed value over time, by quintile.
Sorted by transit system rather than individual station, rates of value creation were
also very highly skewed.Transit systems experiencing the greatest differential rates of change in
total assessed value between treatment and control over the period included those in Phoenix and
Salt Lake City. Those with the lowest differential rates of change between treatment and control
over the 11-year period included those in Denver, Los Angeles, Charlotte, Minneapolis, and San
Diego, as illustrated in the Figure 12.
1 (top 20%) 2 3 4 5 (Lowest20%)
Series1 306.72 5.87 0.93 -0.93 -12.46
-13
37
87
137
187
237
287
337
Diffe
rent
ial R
ate
of C
hang
e (%
) in
Tota
l As
sess
ed V
alue
By S
tatio
n Q
uint
ile
81
Figure 13: Mean differential rates of change in Total Assessed Value by System
Heterogeneity in differential assessed value creation within systems.
Not only did differential rates of total assessed value creation vary significantly across
transit systems, these rates were highly heterogeneous within systems. The same was true for
assessed values of land and improvements, and for changes in folio density. Even along a single
transit line where differential value creation over the 10-year period was very high for some
stations, it was effectively flat or negative for others. The 13-stations studies within Charlotte’s
CATS Blue Line provide an illustrative example. Total assessed value per acre within Transit
Areas of Influence within the 13-station areas increased 146% from 2005 through 2015, whereas
total assessed value per acre increased only 54% within surrounding control areas over the same
period. Total assessed value per acre increased 92% more in Charlotte’s treatment areas than in
control areas. These results were very uneven across individual stations, however. Four of the
Blue Line stations experienced differential assessed value creation per acre of more than 100%
over the period; three station areas experienced differential total assessed valuation increases of
0 50 100 150 200 250
San Diego, SprinterMinneapolis Metro Transit
LA Metro, GoldCharlotte, CATSLA Metro, Expo
Denver, RTDSeattle, SoundTransit Link
Portland, TriMet MaxDallas, DARTNorfolk, Tide
New Jersey TransitHouston, METRO
Salt Lake City, TRAXPhoenix, Valley Metro
82
between 14% and 62% over the period; and the remaining six station areas experienced negligible
or negative differential value creation over the period.
Such variation between station areas was common along many of the subject transit
lines and within most of the transit systems. Only the Red Line in Salt Lake City experienced
uniformly positive differential total assessed valuation growth per acre over the period of study.
The Metro Blue Line in Minneapolis and the RTD Purple Line in Denver each had only a single
station area with negative differential total assessed value creation. A relatively few station areas
experiencing very high rates of differential value creation account for the aggregate average
increase of more than 50% over the period.
This study was concerned with differential rates of assessed value creation between
Transit Areas of Influence (treatment areas) and surrounding control areas. Of the transit systems
studied, Seattle is the outlying case of very high levels of absolute total assessed value increases
per acre both within treatment and control areas, resulting in low differential rates of assessed
value creation per acre. In some cases, values increased significantly faster within treatment areas
than within control areas and surrounding other stations just the opposite was true.
Treatment effect – significance and magnitude
The first mixed model ANOVA estimated the effects of time, treatment, and the
interaction between the time and treatment controlling only for location (transit system) and
various demographic covariates. Total assessed value was regressed on a broad array of
demographic characteristics to identify and eliminate variables with significant collinearity. The
resulting list of demographic (control) covariates includes the following at the census tract-level;
median household income, per capita income, total vehicles per household, percent of dwellings
83
vacant, percent of dwellings owner occupied, cars per owner occupied household, persons per
household, and gross rent burden greater than 30%.
Main and interaction effects of this first mixed ANOVA analysis are as follow:
Main Effects. The main effect of time was significant, F(1.72, 746.48), p = .0.41,
η2partial = .01. This indicates that, controlling for location (transit system) and demographic
covariates, there were significant differences in total assessed value per acre from year to year
between 2005 and 2015. The main effect of group (treatment and control) was significant, F(1,
433) = 4.37, p = .041, η2partial = .01. This indicates that, when controlling for location and
demographic characteristics, there is a significant difference between treatment and control areas,
but that this effect accounted for only approximately 1.0% of the variance in total assessed total
value per acre, exclusive of the interaction of treatment and time (based on the Eta squared [η2
partial] statistic).
Eta squared [η2] measures the proportion of the total variance in a dependent variable that is associated with the membership of different groups defined by an independent variable. Partial eta squared [η2partial] is a similar measure in which the effects of other independent variables and interactions are partialled out. (Richardson, 2011) Table 2: ANOVA Source Table, Total Assessed Value (1), Between Subjects
Source df F p η2partial
Treatment 1.00 4.37 .037 .01 Error 433.00 - - -
Covariates. Significant covariates included transit system, F(130, 746.48) = 9.32, p < .001,
η2partial = .22, per capita income, F(1.72, 746.48) = 15.68, p < .001, η2partial = .04), and vehicles per
household, F(1.72, 746.48) = 6.63, p < .002, η2partial = .02), These eta squared-partial results suggest
that prior to controlling for individual station-specific characteristics, 22% of the differential rate of
change in total assessed value per acre over time (11-years) can be explained by the unique
locational (perhaps “market”) characteristics associated with each transit system. An additional 4%
84
of this variation is explained by variation in per capita income levels at the census block-level, and
2% by the number of vehicles per household.
Interaction effect. There was a significant interaction between time and treatment
group, F(1.72, 746.48) = 10.31, p < .001, η2 partial = .02, as reflected in Table 3, indicating that
there were
significant differences in rates of change in total assessed value per acre between treatment and
control groups over time, while controlling only for location (transit system) and demographic
covariates; and that the “treatment effect” accounted for approximately 2% of the variance in
differences in total assessed value per acre over time (difference-in-differences).
85
Table 3: ANOVA Source Table, principal treatment effect, Total Assessed Value (1)
Source df F p η2partial Time 1.72 3.39 .041 .01 Time x Treatment 1.72 10.31 < .001 .02
Time x Median Household Income
1.72 2.79 .070 .01
Time x Per Capita Income
1.72 15.68 < .001 .04
Time x Cars per Household
1.72 6.63 .002 .02
Time x Percent Houses Vacant
1.72 0.87 .404 .00
Time x Percent Houses Owner Occupied
1.72 1.21 .296 .00
Time x Cars per Owner Occupied Household
1.72 0.50 .578 .00
Time x Household Size
1.72 2.77 .072 .01
Time x Rent Burden >30%
1.72 0.51 .573 .00
Time x System 130.00 9.32 < .001 .22
Error 746.48 - - -
Station characteristics - significance and magnitude
Following estimation of the significance and magnitude of the principal treatment
(proximity to LRT stations) effect on differential rates of total assessed value creation, station-
specific characteristics were evaluated for association with variation in treatment effect between
stations. Station-specific characteristics fell into two general classes. The first of these consisted
of physical and function characteristics of stations such as station design, (locational) typology,
(walk-and-ride or park-and-ride) character, and number of transit agency-provided parking
86
spaces. The second class of characteristics was comprised of a) transit agency goals and
objectives driving planning/development of subject transit line and station areas; b) value
creation strategies employed in connection with transit line/station planning and development; c)
value capture mechanisms employed as part of the infrastructure funding strategy; and d)
significance or importance of value capture strategy to various aspects of project success.
Specific questions associated with each of these inquiries are identified in the survey instrument
provided as Appendix I.
Main and interaction effects of this second mixed ANOVA analysis, as reflected in
Table 4, are as follow:
Main Effects. The main effect of time was not significant, F(1.72, 718.04) = 47.31, p
= .190, η2partial = .02. This indicates that, when controlling for all other effects, there were not
significant differences in assessed total value per acre from year to year between 2005 and 2015.
The main effect of group (treatment and control) was significant, F(1, 418) = 4.39, p = .037,
η2partial = .01. This indicates that, when controlling for all other effects, there is a significant
difference between treatment and control areas, but that this effect accounted for only
approximately 1.0% of the variance in total assessed total value per acre, exclusive of the
interaction of treatment and time.
Table 4: ANOVA Source Table, Total Assessed Value (2), Between Subjects
Source df F p η2partial
Treatment 1.00 4.39 .037 .01 Error 418.00 - - -
Covariates. Significant covariates included per capita income, F(1.72, 718.04) =
11.01, p < .001, η2partial = .03), importance of public-private value creation strategies to the
success of transit-influenced or transit-oriented development, F(1.72, 718.04) = 5.92, p = .005,
87
η2partial = .01, importance of public-private value creation strategies to the success of transit
infrastructure investments and projects, F(1.72, 718.04) = 6.06, p = .004, η2partial = .01, station
design, F(5.15, 718.04) = 2.48, p = .029, η2partial = .02 and station system, F(22.33, 718.04) =
7.31, p < .001, η2partial = .19. This suggests that per capita income, importance of public-private
value creation strategies to the success of transit-influenced or transit-oriented development,
importance of public-private value creation strategies to the success of transit infrastructure
investments and projects, station design, and station system significantly associated with assessed
total value over time, account for variance as indicated by the partial eta squared statistic for
each.
Interaction effect. There was a significant interaction between time and treatment
group, F(1.72, 718.04) = 10.42, p < .001, η2 partial = .02, indicating that there were significant
differences in rates of change in assessed total value per acre between the treatment and control
groups over time, while controlling for the covariates, accounting for approximately 2% of the
variance in differences in assessed total value per acre (i.e., difference-in-differences). Figure 13
reflects estimated marginal means of values of total assessed value within treatment and control
groups over time adjusted for all covariates.
88
Figure 14: Estimated Marginal Means, Total Assessed Value
Estimated Marginal Means
Estimated marginal means as reflected in Figures 13, 17, 21, and 24 presented for
each of the four fully controlled models reported in this study represent the estimated mean
values for each subject dependent variable (in each year), controlled for each of the covariates in
each model. These are the means of dependent variables estimated using the mean values of each
of the covariates. Actual mean values for each dependent variable are estimated
(adjusted) based on the mean values of the (17) covariates in each model.
89
Table 5: ANOVA Source Table, Station Characteristics, Total Assessed Value (2), Within-
Subjects
Source df F p η2partial Time 1.72 1.69 .190 .00 Time x Treatment 1.72 10.42 .000 .02 Time x Median Household Income 1.72 2.68 .078 .01 Time x Per Capita Income 1.72 11.01 .000 .03 Time x Cars per Household 1.72 1.58 .209 .00 Time x Percent Houses Vacant 1.72 0.95 .375 .00 Time x Percent Houses Owner Occupied 1.72 0.66 .493 .00 Time x Cars per Owner Occupied Household 1.72 0.77 .447 .00 Time x Household Size 1.72 0.33 .687 .00 Time x How significant/important was value capture strategy to financial viability or project success? 1.72 1.22 .291 .00 Time x How important are public-private value creation strategies to the success of transit-influenced or transit-oriented development? 1.72 5.92 .005 .01 Time x How important are public-private value creation strategies to the success of transit infrastructure investments and projects? 1.72 6.06 .004 .01 Time x How important is value capture to the success of transit infrastructure investments and projects? 1.72 0.59 .531 .00 Time x Rent Burden >30% 1.72 0.40 .637 .00 Time x Parking Spaces 1.72 0.47 .598 .00 Time x Station Design 5.15 2.48 .029 .02 Time x Station Typology 10.31 0.42 .943 .01 Time x System 22.33 7.31 .000 .19 Time x Character
1.72
0.66
.496 .00
Error 718.04 - - -
90
Supplemental Analyses
Assessed Value-Land
Prospectively differential rates of change in the assessed values of land (as opposed to
that of improvements) over time is of some interest due to theory that economic benefits of
mobility and accessibility resulting from proximity to transit stations (or that associated with
other beneficial public infrastructure investment) will be capitalized into the market value of land
in particular. The theoretical assumption that such benefits will accrue to land values per se is
much of the rationale supporting land value taxation (Bryson, 2011; George, 1879).
Aggregate totals of assessed value-land within each geographic treatment and control
area were divided by the acreage in each area resulting in values per acre to facilitate level
comparison. Assessed value-land within 229 subject treatment areas rose from an average (mean)
of $145,805 per acre in 2005 to $290,895 per acre in 2015, an increase of 100% over the period,
or an average annual rate of 9.95%. Within corresponding control areas, total assessed value per
acre rose from an average of $156,783 per acre to $261,357 per acre, increasing at an average
annual rate of 6.67%. Across all 229 stations, total assessed valuation per acre within transit
treatment areas increased at an average rate of 3.28% per year faster than that within control
areas as illustrated in Figure 14. These results were highly varied across transit lines and stations,
however.
91
Figure 15: Mean Assessed Value-Land per acre
Although differences between aggregate rates of assessed land value creation over
time were only modest, even these were highly varied across stations. The average aggregate rate
of differential assessed land value creation associated with proximity to LRT stations within the
20% of transit systems exhibiting the highest rates of increase in assessed value-land within
treatment areas over time experienced an average increase of 191.7% over the period, or an
average annual rate of 19.17%. Whereas stations within the 20% of transit systems exhibiting the
lowest rates of increase in assessed value-land within treatment areas over time experienced an
average decrease of 28.24% over the period, or an average annual rate of negative 2.82%, as
Figure 16: Differential (mean) rates of change in assessed land value over time, by quintile. Sorted by transit system rather than individual station, differential rates of value
creation were much less skewed (and all positive), but still highly varied. Transit systems
experiencing the greatest differential rates of change in total assessed value between treatment
and control over the period included those in Salt Lake City and Denver. Those with the lowest
differential rates of change between treatment and control over the 11-year period included those
in Charlotte, Los Angeles, Minneapolis, and San Diego, as illustrated in Figure 16.
1 (top 20%) 2 3 4 5 (Lowest20%)
Series1 191.70 5.91 1.00 -1.17 -28.24
-5.00
45.00
95.00
145.00
195.00
Diffe
rent
ial R
ate
of C
hang
e (%
) in
Asse
ssed
Val
ue
(Lan
d)
Quintile
93
Figure 17: Mean differential rates of change in Assessed Land Value by System
Mixed model ANOVA was used to determine the effects of treatment, time, and the
interaction between the two on assessed value–land as the dependent variable. Main and
interaction effects of this analysis are as follow.
Main Effects. The main effect of time was not significant, F(1.69, 704.34) = 2.25, p =
.116, η2partial = .01. See Table 6. This suggests that, when controlling for all other effects, there
were no significant global differences in assessed land values per year per acre during the time
period of 2005 to 2015. The main effect of group was not significant, F(1, 418) = 0.05, p = .832,
η2partial = .00. This indicates that, when controlling for all other effects, there is not a significant
difference between treatment and control areas.
Table 6: ANOVA Source Table for Assessed Land Value per Acre, Between-Subjects
Source SS df MS F p η2partial Treatment 18522372670.00 1.00 18522372670.00 0.05 .832 .00 Error 171467944300000.00 418.00 410210393000.00 - - -
Covariates. Significant interactions with covariates included per capita income, F(1.69,
704.34) = 3.44, p = .040, η2partial = .01, importance of public-private value creation strategies to the
0 20 40 60 80 100 120
San Diego, SprinterMinneapolis Metro Transit
LA Metro, ExpoCharlotte, CATS
Dallas, DARTSeattle, SoundTransit Link
Norfolk, TideNew Jersey Transit
Portland, TriMet MaxHouston, METRO
Phoenix, Valley MetroDenver, RTD
Salt Lake City, TRAX
Average Differential Rate of Change (%) in Assessed Land Values
94
success of transit-influenced or transit-oriented development, F(1.69, 704.34) = 5.14, p = .009,
η2partial = .01, and station system, F(21.1, 704.34) = 5.56, p < .001, η2partial = .15. This indicates that
per capita income, importance of public-private value creation strategies, and type of station
system significantly influenced assessed land value over time, accounting for up to 15% (in the
case of station system) of the variation in assessed land value.
Interaction effect. There was a significant interaction between time and treatment group,
F(1.69, 704.34) = 4.34, p = .019, η2 partial = .01, indicating that there were significant differences in
rates of assessed land values per acre between the treatment and control groups over time, while
controlling for the covariates (i.e., difference-indifferences). The partial eta squared coefficient
indicates that approximately 1% of the variation in differences in assessed land value can be
attributed to this interaction. The interaction effect was not significant in any individual year. The
following figure reflects estimated marginal means of values of assessed value–land within
treatment and control groups over time adjusted for all covariates.
95
Figure 18: Estimated Marginal Means, Assessed Value – Land
96
Table 7: ANOVA Source Table, Station Characteristics, Assessed Land Value, Within-Subjects
Assessed Value-Improvements
Source df F p η2partial Time 1.69 2.25 .116 0.01 Time x Treatment 1.69 4.34 .019 0.01
Time x Median Household Income 1.69 0.96 .370 0.00
Time x Per Capita Income 1.69 3.44 .040 0.01
Time x Cars per Household 1.69 0.48 .585 0.00
Time x Percent Houses Vacant 1.69 0.20 .781 0.00
Time x Percent Houses Owner Occupied 1.69 0.57 .536 0.00
Time x Cars per Owner Occupied Household 1.69 0.99 .360 0.00
Time x Household Size 1.69 0.66 .493 0.00
Time x How significant/important was value capture strategy to financial viability or project success?
1.69 0.61 .518 0.00
Time x How important are public-private value creation strategies to the success of transit-influenced or transit-oriented development?
1.69 5.14 .009 0.01
Time x How important are public-private value creation strategies to the success of transit infrastructure investments and projects?
1.69 2.96 .062 0.01
Time x How important is value capture to the success of transit infrastructure investments and projects?
1.69 0.52 .564 0.00
Time x Rent Burden >30% 1.69 0.34 .674 0.00
Time x Parking Spaces 1.69 0.64 .501 0.00
Time x Station Design 5.06 1.72 .128 0.01
Time x Station Typology 10.11 1.56 .115 0.02
Time x System 21.91 5.56 .000 0.15
Time x Character 1.69 0.85 .413 0.00
Error 704.34 - - -
97
As for assessed value-land, aggregate totals of assessed value-improvements within
each geographic treatment and control area were divided by the acreage in each area resulting in
values per acre to facilitate level comparison.
Assessed value-improvements within 229 subject treatment areas rose from an
average (mean) of $310,779 per acre in 2005 to $681,866 per acre in 2015, an increase of 119%
over the period, or an average annual rate of 11.94%. Within corresponding control areas, total
assessed value per acre rose from an average of $241,687 per acre to $425,216 per acre,
increasing at an average annual rate of 7.59%. Across all 229 stations, total assessed valuation
per acre within transit treatment areas increased at an average rate of 4.35% per year faster than
that within control areas, as illustrated in Figure 18.
Figure 19: Mean Assessed Improvement(s) Value per acre
These values were highly varied across stations. The average aggregate rate of
differential assessed improvement value creation associated with proximity to LRT stations
within the 20% of transit systems exhibiting the highest rates of increase in assessed
improvement(s) value within treatment areas over time experienced an average increase of
223.93% over the period, or an average annual rate of 22.39%. Whereas stations within the 20%
Source df F p η2partial Time 1.99 2.05 .130 .01 Time x Treatment 1.99 6.74 .001 .02 Time x Median Household Income 1.99 0.31 .735 .00 Time x Per Capita Income 1.99 7.09 .001 .02 Time x Cars per Household 1.99 2.08 .126 .01 Time x Percent Houses Vacant 1.99 2.33 .098 .01 Time x Percent Houses Owner Occupied 1.99 0.19 .824 .00 Time x Cars per Owner Occupied Household 1.99 0.40 .667 .00 Time x Household Size 1.99 3.00 .050 .01 Time x How significant/important was value capture strategy to financial viability or project success? 1.99 0.61 .546 .00 Time x How important are public-private value creation strategies to the success of transit-influenced or transit-oriented development? 1.99 4.50 .011 .01 Time x How important are public-private value creation strategies to the success of transit infrastructure investments and projects? 1.99 1.07 .345 .00 Time x How important is value capture to the success of transit infrastructure investments and projects? 1.99 0.09 .916 .00 Time x Rent Burden >30% 1.99 0.99 .372 .00 Time x Parking Spaces 1.99 0.51 .602 .00 Time x Station Design 5.98 1.12 .348 .01 Time x Station Typology 11.96 1.36 .182 .02 Time x System 25.91 2.29 .000 .07 Time x Character 1.99 0.67 .510 .00 Error 833.08 - - -
Qualitative results
Planners and/or managers of the 12 of 14 subject transit agencies responding to the
survey (instrument is included below as Appendix I) provided an opportunity to investigate their
perspectives regarding various aspects of value capture, through four open-ended questions.
The first of these was: “How has the importance of value capture changed over time?”
The consensus response was that as funding sources in general, and federal funding in particular,
have become more difficult to secure, local (non-state) funding has become and will continue to
108
be increasingly important. Value capture is a central component of local funding strategies and
may become a determinant of success for future projects. The following comment was typical:
In the early days of development … funding was ‘plentiful’ or at least less competitive and more available at the federal, state and local levels. Given the funding climate at all [three] levels now, there is significant pressure for local agencies and project sponsors to bring more funds to the table and [value capture] - at least as a concept - is rising to a position much higher on the list of possible sources. Financial responsibility is [now] much more at [the] local level. The second of four questions asked of transit agency planners and managers was:
“What do you perceive to be the greatest barriers to realizing transit-related value capture?”
Responses to this question fell into three categories; 1) limited independent authority of transit
agencies to undertake and impose value capture strategies and/or limited (state) statutory
authorization of specific value capture tools; 2) political and/or public policy tension, particularly
with respect to TIF, regarding “diversion” of tax revenues that could otherwise be expended on
other (non-transit) projects; and 3) inadequacy of “business case” arguments that transit
infrastructure investments are, in fact, producing the “surplus” value to be captured to offset that
investment.
The third survey question asked of transit agency planners and managers was: “What
lessons has your experience provided with respect to transit infrastructure related value creation
and value capture?” Although there was some variation in perspective here, the dominant
message was that education of policy makers, stakeholders, and the public at large with respect to
the benefit and equity of value capture is a time-consuming effort requiring significant “front
end” commitment of resources. Project-specific comments included the following:
Making the business case to redirect funds from the municipal general fund remains a challenge. Growth in property taxes, as an example, is usually already embedded in the municipal budget and revenue growth curve (i.e. revenues are "spoken for") long before a transit department can try and make the case for value capture if it is not part of the project predevelopment strategy.
109
The question: “Do you have any suggestions or recommendations for implementing
future transit-related value creation and/or value capture projects?” resulted in recommendations
that included starting the process of engagement with prospective public and private sector
partners very early in the pre-development process and incorporating value capture strategies
(and revenue) in project pre-development financial planning.
All surveyed transit agencies articulated goals and responded to inquiries with
responses consistent with a need to better understand conditions precedent to value creation and
those factors contributing to variation in development success and value creation in proximity to
transit stations. Transit planners and managers are increasingly interested in understanding the
value creation process, particularly with a view towards facilitating greater and more productive
third-party engagement, and engaging earlier and more affirmatively in value creation strategies.
Summary of findings
As hypothesized, differential rates of assessed value creation varied widely across
transit systems and individual stations. Transit Areas of Influence (TAIs) in the top 20% of
differential value creation experienced average annual growth rates 30.67% faster than that of
control, whereas TAIs in the bottom quintile experienced negative average annual differential
growth rates (1.25% less than those within control areas). Significantly positive differential
value creation was concentrated within a small number of transit systems and within a relatively
few stations along several of the lines studied.
Differential rates of assessed value creation are found to accrue disproportionately to
improvements (and to folio density) rather than to land. In the aggregate, assessed
improvement(s) values grew 4.35% faster in treatment areas than in control, whereas assessed
land values grew only 3.28% faster within treatment areas. The extent to which treatment
explained variation in differential value creation was approximately twice that for assessed
110
improvement(s) value and that for assessed land value; approximately 2% compared to
approximately 1%. Additionally, although the treatment-time interaction effects were significant
over the entire period for assessed improvement values, the effect was not significant in any
individual year for assessed land value.
Significant covariates included transit system (location), accounting for 19% of the
differential rate of change in total assessed value per acre over time; per capita income (at the
census block-level), accounting for an additional 3% of variation; station design, accounting
2% of variation; transit agency perceptions of public-private value creation strategies as
important 1) to the success of transit infrastructure investments and projects, and 2) to the
success of transit-influenced or transit-oriented development, accounted for an additional 1% of
variation each. Significantly positive differential value creation occurred predominantly near
at-grade stations.
Of numerous demographic covariates evaluated, only per capita income and vehicles
per household were significant, predicting 4% and 2%, respectively, of differential rates of
change over time (the interaction of covariate and time) between treatment and control groups.
Implications for literature
The findings that the extent of differential assessed value creation varies significantly
from one market, transit line, or station to another is consistent with that variation on consumer’s
marginal willingness to pay for increased mobility, enhanced accessibility, and/or other related
amenities such as those associated with TOD. Results of this study both underscore the extent of
that variability and reveal variation in rates of value creation over a significant period of time.
This study quantifies various aspects of differential value creation in terms of assessed valuation,
the lingua franca of municipal and ad valorem tax revenue-based finance, rather than in terms of
the more theoretical economic framework of consumers’ marginal willingness to pay. Other
111
implications with respect to academic literature are discussed below under opportunities for
future research. These include the need for more complete understanding of 1) the economic,
market, and sub-market factors that allow robust differential value creation in some places, but
flat or negative outcomes in others; 2) factors, perhaps including institutional practices or
constraints, result in value creation becoming (apparently) disproportionately assessed to values
of improvements as opposed to land; and 3) forces that allowed rates of differential value
creation to be so much greater within some markets (transit systems) than others over the same
A second analysis was performed on the 200 of 229 station areas for which there were
no missing annual assessed valuation data. Regression analysis was performed on panel data
constructed based on a different research design premise than that underpinning the ANOVA
analysis reported above. Each of the (502.66 acre) Transit Areas of Influence (treatment areas)
within ½-mile of station area centroids was considered in its entirety, regardless of whether or not
that treatment area encroached into and overlapped one or more adjoining Transit Areas of
Influence. Whereas Thiessen polygons and a GIS union method were previously employed to
eliminate overlapping treatment areas so as to avoid double counting of folio values within
treatment areas (described in Research Methods and Design, Chapter 3 and within Appendix III),
no such “cropping” of treatment areas was undertaken for this alternative (panel regression)
analysis; this revised definition of treatment areas includes every folio within ½-mile of station
centroids.
The following figure illustrates the overlapping ½-mile radius Transit Areas of
Influence surrounding the Farmdale and Expo/La Brea stations on the La Brea Line in Los
Angeles. The small dots represent centroids of all taxable folios within TAIs/treatment areas.
Values for folios falling within both station area TAIs are included within treatment for each
station.
113
Figure 25: Overlapping Transit Areas of Influence and Treatment Areas
Although this analysis considers a slightly different (reduced) dataset and was based
on a different definition of treatment geographies, overall results were similar to those reported
under ANOVA above. Aggregate total assessed value within treatment areas increased 146% over
the11-years of measurement from 2005 through 2015, whereas that value within control areas
increased only 82% over the period. Aggregate total assessed value per acre within treatment
areas increased at an average annually compounded rate of 3.21% faster than that within control
areas, or a total of 63% over the period.
114
Table 12: Differential Rates of Total Assessed Value Growth per Acre, by Station Quintile
As in the previous analysis, however, differential rates of value increase were highly
varied across stations. The aggregate differential rate of assessed value creation per acre within
the “top” quintile of stations was more than twice that within the second quintile; and the rate was
negative within the lowest quintile. As anticipated, differential rates of value creation were
somewhat higher across many stations in this alternative analysis as a result of some high-value
properties (office or condominium towers, for example) in proximity to station areas being
captured within two or more treatment areas.
Figure 26: Differences Between (Compounded) Annual Rates of AV Creation Within Treatment and Control Areas, by Quintile
115
As in the analysis above, differential rates of assessed value creation per acre were
also highly varied across transit systems. Differences between compounded annual rates of value
creation within treatment and control areas ranged from -1.16% to 7.82% (3.21% average across
all stations, regardless of system). Note however that differential rates of value creation were
uncorrelated with absolute levels of assessed value per acre or value creation. Assessed values per
acre, and year over year increases in some densely urban areas of Seattle, for example, were
among the highest in the study. Yet differential rates of value creation within treatment areas grew
at a slightly lower rate than in surrounding control areas where assessed values per acre grew at
even higher rates.
Table 13: Differential Rates of Total Assessed Value Growth per Acre, by Transit System
116
Figure 27: Differences Between (Compounded) Annual Rates of AV Creation
Within Treatment and Control Areas, by Transit System
Rates of differential total assessed value per acre were highly varied within (along)
specific transit lines as well as across all transit systems and stations. Results from the Blue Line
in Charlotte, North Carolina are provided as illustrative example.
Table 14: Differential Rates of Total Assessed Value Growth/Acre by Charlotte Blue Line Station
Differences between compounded annual rates of value creation within treatment and
control areas within the 13-stations studied along the CATS Blue Line ranged from -3.23% to
9.94% (5.01% on average). Seven of the 13-station experienced positive differential rates of value
117
creation per acre over the study period, while the other six experienced lower rates of value
creation within treatment areas than within surrounding control areas.
Figure 28: Differences Between (Compounded) Annual Rates of AV Creation Within Treatment and Control Areas Along Charlotte’s Blue Line by Transit System
Highly varied outcomes resulted notwithstanding that similar levels of regulatory
entitlement and infrastructure investment were provided within each station area and Transit Area
of Influence. Transit-Oriented Development has been encouraged in all station areas, and similar
development incentives have been provided. Charlotte’s policy makers and planners within both
the Charlotte-Mecklenburg Planning Department and the Charlotte Area Transit System (CATS)
are sensitive to the extent of variation in market response to investment in Blue Line stations along
the corridor and continue to explore opportunities for encouraging more growth and development
near stations where market response has been modest and/or where there has been little or no
development activity.
Panel Regression
Differential rates of total assessed value growth per acre, by station per year, were
regressed on a variety of prospectively explanatory variables to estimate the extent to which they
118
might be associated with differential rates of assessed value growth. Independent variables were
comprised of transit system, station characteristics (design, typology, character), transit agency
goals and objectives, value creation and value capture strategies, the relative importance to transit
agencies of various objectives and outcomes, and a variety of demographic characteristics, all
previously described. Given that many of the independent (prospectively explanatory) variables
are time-invariant with respect to stations, random-effects panel regression was employed (A
Kohler & A Kreuter, 2012). Practical experience, intuition, and results presented herein all
suggest that differences across stations, lines, and transit systems (metropolitan markets)
unaccounted for in these analyses, have influence on the rate of assessed value creation per acre.
Consistent with ANOVA results reported above, very few (perhaps surprisingly few) of the
prospective predictor variables were of statistical significance.
Prior to performing regression analyses, a paired t-test was performed to confirm
there was a statistically significant mean difference between total assessed values per acre within
treatment areas and those within control areas. As previously reported, mean total assessed value
per acre was higher within treatment areas than within control areas. Mean total assessed value
per acre within treatment areas, across all periods, was $734,893 (95% CI, $678,442 to $791,343),
compared to $551,469 within control areas (95% CI, $529,857 to $573,082). There was a
statistically significant mean difference of $183,423 per acre (95% CI, $137,343 to $229,503);
t(2199) = 7.8060, p < 0.0005.
From 2005 through 2015, mean total assessed value per acre increased from
$451,185 to $1,108,469 with treatment areas, and increased from $392,200 to $715,636 within
control areas. The difference between mean total assessed value per acre within in treatment areas
and that within control areas increased from $58,985 in 2005 to $392,833 in 2015.
119
A second paired t-test was performed to determine whether there was a statistically
significant mean difference between annual growth rates (in total assessed values per acre) within
treatment areas and those within control areas. As previously reported, growth rates were higher
within treatment areas than within control areas in the aggregate. The mean annual growth rate
within treatment areas was 10.95% (95% CI, 9.21% to 12.72%), compared to 6.91% within
control areas (95% CI, 6.01% to 7.81%). There was a statistically significant mean difference of
4.04% per acre (95% CI, 2.52% to 5.57%); t(1999) = 5.1997, p < 0.005.
Differential rates of total assessed value growth per acre by station were regressed
on time (year) to evaluate the extent to which any particular year (or all years) were significant
predictors of variation in growth rates. As in all following models, assessed value per acre is
differenced at the station level. The outcome of interest is the relative difference in the rate of
change of assessed valuation per acre within treatment areas compared to that within control
areas. For this reason, the regression constant is a material value rather than a nuisance parameter
(as it might be in some social science models). In the following model, for example, the constant
was significant, p = 0.031, and the coefficient was positive, reflecting a significant effect of
treatment. Between 2005 and 2015, only 2008 was significant, p = 0.033. Differential rates of
growth were also regressed on the years-in-service of stations (elapsed time since commencement
of service), resulting in no statistical significance.
120
Table 15: Panel Regression Differential Rates Total Assessed Value Growth/Acre on Year (only)
Random-effects GLS regression Number of obs = 2,000 Group variable: ID Number of groups = 200 R-sq: Obs per group:
within = 0.0000 min = 10 between = 0.0000 avg = 10
overall = 0.0047 max = 10 (Robust Standard. Error adjusted for 200 clusters in ID) Wald chi2(9) = 17.18 corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.046 Differential AV Growth Rate Coef. R. Std. Err. z P>z [95% Conf. Interval] YEAR
Walk and Ride 0.0274400 0.0325725 0.84 0.400 -0.036401 0.0912809 _cons 0.0078890 0.0317317 0.25 0.804 -0.054304 0.070082 sigma_u 0.07320754 sigma_e 0.34501818 rho 0.04308262 (fraction of variance due to u_i)
Differential rates of total assessed value growth per acre by station were regressed
(solely) on transit agencies’ primary objectives driving planning/development of subject transit
line and station areas (as previously defined, and as reported through transit agency survey
responses); none of which were significant.
123
Table 17: Panel Regression Differential Rates of Total Assessed Value Growth per Acre on Reported Transit Planning and Development Objectives (only)
Random-effects GLS regression Number of obs = 2,000 Group variable: ID Number of groups = 200 R-sq: Obs per group: within = 0.0000 min = 10 between = 0.3446 avg = 10 overall = 0.0499 max = 10
(Robust Standard. Error adjusted for 200 clusters in ID) Wald chi2(7)
Differential rates of total assessed value growth per acre by station were regressed
(solely) on transit agencies’ perceived significance/importance of various value creation and
capture strategies to financial viability or project success, TOD development success, and the
success of transportation infrastructure investments (as reported through transit agency survey
responses). Again, the outcome of interest is the relative difference in the rate of change of
assessed valuation per acre within treatment areas compared to that within control areas. In this
model, the regression constant was significant, p = 0.009, and the coefficient was positive,
reflecting a significant effect of treatment. None of independent covariates, however, were
statistically significant predictors of differential rates of value creation.
125
Table 19: Panel Regression Differential Rates of Total Assessed Value Growth per Acre on Reported Transit Agency Perceptions of the Importance of Various Value Creation and Capture Strategies to Financial Viability or Project Success (only)
Random-effects GLS regression Number of obs = 1,810 Group variable: ID Number of groups = 181 R-sq: Obs per group:
within = 0.0000 min = 10 between = 0.0700 avg = 10
overall = 0.0077 max = 10 (Robust Standard. Error adjusted for 181clusters in ID) Wald chi2(11) = 40.91 corr(u_i, X) = 0 (assumed) Prob > chi2 = 0 Differential AV Growth Rate Coef. R. Std. Err. z P>z [95% Conf. Interval] VC_FINIMPORTANCE
Slightly Important 0.0205275 0.0280356 0.73 0.464 -0.0344213 0.0754763 Moderately Important -0.0116708 0.0164685 -0.71 0.479 -0.0439484 0.0206068
Significantly Important -0.0217834 0.018231 -1.19 0.232 -0.0575156 0.0139487 Very Important 0.0123983 0.0265667 0.47 0.641 -0.0396714 0.0644681
PPE_TOD_IMPORTANCE
Slightly Important 0.0272859 0.0193435 1.41 0.158 -0.0106267 0.0651984 Moderately Important -0.0336873 0.0192787 -1.75 0.081 -0.0714728 0.0040982
Significantly Important 0.0105012 0.0165517 0.63 0.526 -0.0219396 0.042942 Very Important -0.0130826 0.0157669 -0.83 0.407 -0.0439851 0.0178199
PPE_TRANS_IMPORTANCE
Moderately Important 0.0172637 0.0147567 1.17 0.242 -0.011659 0.0461864 Significantly Important -0.0132616 0.0194653 -0.68 0.496 -0.051413 0.0248897
sigma_u 0.018191 sigma_e 0.1755852 rho 0.0106194 (fraction of variance due to u_i)
Differential rates of total assessed value growth per acre by station were regressed
(solely) on value capture strategies employed in connection with transit line/station planning and
development (joint development, negotiated exactions, special assessments, naming rights, or land
value taxation, as reported through transit agency survey responses); none of which were
126
significant. Once again, however, the regression constant was significant, p = 0.009, and the
coefficient was positive, reflecting a significant effect of treatment itself.
Table 20: Panel Regression Differential Rates of Total Assessed Value Growth per Acre on Reported Employment of Various Value Capture Strategies (only)
Random-effects GLS regression Number of obs = 2,000 Group variable: ID Number of groups = 200 R-sq: Obs per group: within = 0.0000 min = 10 between = 0.0472 avg = 10 overall = 0.0068 max = 10 (Robust Standard. Error adjusted for 200 clusters in ID) Wald chi2(4) = 5.66 corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.2264 Differential AV Growth Rate Coef. R. Std. Err. z P>z [95% Conf. Interval] VC_JOINTDEVELOPMENT -0.0492563 0.040859 -1.21 0.228 -0.1293384 0.0308258 VC_EXACTIONS 0.0118525 0.0256562 0.46 0.644 -0.0384327 0.0621377 VC_SADS -0.0526231 0.0349949 -1.50 0.133 -0.1212119 0.0159656 VC_NONE -0.0750624 0.0477282 -1.57 0.116 -0.1686078 0.0184831 _cons 0.1021178 0.0474861 2.15 0.032 0.0090467 0.1951889 sigma_u 0.0749699 sigma_e 0.338917 rho 0.0466488 (fraction of variance due to u_i)
The full random-effects panel regression model was specified employing all
prospective explanatory variable not dropped from regression due to collinearity. The time
specification included the 11 years from 2005 through 2015. Independent variables included
transit system (S_SYSTEM), years in service, station design (as indicated in following table, with
“underground” specified as base condition), station typology (as indicated, with “campus/other”
specified as base condition), station character (with “park and ride” specified as base condition),
transit agencies’ primary objectives driving planning/development of subject transit line and
station areas (as previously defined), value creation strategies employed by transit agencies in
connection with transit line/station planning and development (as previously defined), specific
127
value capture strategies employed, transit agencies’ perceived significance/importance of value
creation and capture strategies to financial viability or project success, TOD development success,
and the success of transportation infrastructure investments, and various demographic
characteristics (previously described).
Statistically significant independent variables with the full model included “At Grade
Station Design” (positive coefficient relative to Underground Station Design, p = 0.019), “Urban
Neighborhood Typology” (negative coefficient relative to Campus/Other, p = 0.001), “Serving
anticipated residents and workers drawn to new development near transit stations” as a principal
transit agency planning/development goal (negative coefficient, p = 0.000, all other transit agency
goals were dropped due to collinearity or lack of response), and providing “land use and zoning
entitlements” as enticement for development as part of a value creation strategy (negative
coefficient, p = 0.015). Of demographic covariates including median household income, rent
burden, home ownership rate, and cars per household, only “rent burden” (those spending 30
percent or more of income on housing costs) was significant with a negative coefficient for
differential value creation, p = 0.016.
Given the limited extent of statistical significance across various station and transit
agency characteristics, collinearity between a number of variables (such as value creation
initiatives with other factors), the extent to which other prospectively explanatory values appear to
absorb the effect of transit agency (S_SYSTEM/metropolitan market), for example, significant
cofounding between independent variables is assumed.
128
Table 21: Full Regression Model: Panel Regression Differential Rates of Total Assessed Value Growth per Acre on All (non-dropped) Independent Variables
Random-effects GLS regression Number of obs = 1,800 Group variable: ID Number of groups = 180 R-sq: Obs per group:
within = 0.0000 min = 10 between = 0.3966 avg = 10
overall = 0.0435 max = 10 (Robust Standard. Error adjusted for 180 clusters in ID) Wald chi2(29) = 7914.93 corr(u_i, X) = 0 (assumed) Prob > chi2 = 0 Differential AV Growth Rate Coef. R. Std. Err. z P>z [95% Conf. Interval] S_SYSTEM -0.0132336 0.0195175 0.68 0.498 -0.0514872 0.02502 YRS_in_SERVICE 0.0004978 0.0011325 0.44 0.660 -0.0017219 0.0027174 S_DESIGN
Real estate is comprised of highly heterogeneous bundles of goods rendered unique
by location. Real estate markets and submarkets are non-uniform, complex, and subject to
significant fluctuation across time and space. Real estate markets are cyclical, but not coincident
across classes of real property. This study’s results suggest that market forces defined and
influenced by factors outside the scope of those considered herein are responsible for greater
fluctuation in differential value creation than that explained by modeled variables. Not the least
of these may have been the overarching impacts and market distortions imposed by the Great
Recession during the period of study. Market responses to stimuli such as transit infrastructure
investment are complex and vary across time and from place to place. Some of the many
potential limitations affecting both generalizability and validity of this study are as follow.
Non-random determination of transit line corridors and alignments, station
locations and design, and institutional factors
A truly experimental research design requires random assignment of treatment to
units of analysis. A concern with this quasi-experimental design is that assignment to treatment
(e.g. design, entitlement, funding, and development of the transit corridors and location-specific
stations) is entirely nonrandom. These decisions and outcomes result from political processes,
economic, financial, market and engineering considerations and constraints, all of which are non-
random.
Data quality
The quantitative analysis reflected herein incorporates three distinct datasets. These
include 1) assessed valuation data comprising annual folio counts, assessed value of land,
assessed value of improvements, and total assessed value, all expressed per acre within treatment
130
and control areas from 2005 through 2015; 2) station-specific descriptive and quantitative data
derived from surveys sent to senior transit system planners and managers; and 3) annual
American Community Survey demographic data from 2010. Issues related to each of these
datasets contributing to potential concerns regarding bias and/or validity are addressed as
follows. The most significant of these concern the assessed valuation data itself, and the way in
which it has been organized and compiled.
Tax assessment and methodology
Assessed valuation data can be considered “murky,” particularly as reflections of
market value in real time. There are several reasons for such concerns. Local governments assess
the value of real and intangible property as the basis for the levy of taxes on which they depend
for revenue. Tax assessors consider comparable sales and other data to estimate market value as
the basis for assigning folio-specific assessed values for tax purposes. Regardless of how dynamic
(or volatile) underlying real estate markets may be, assessed values are updated only annually,
usually as of the first of January. Actual reevaluation and appraisal of individual properties may
only occur at intervals of four to eight years or longer. Therefore, assessed values are not
extremely sensitive or responsive to changes in underlying market value in real time. Assessed
values lag market values both in expanding and contracting markets. Matters are complicated
further by the fact that assessed values are influenced by political and policy considerations which
may discount market values and/or attempt to moderate the effects of market volatility.
Market value is itself essentially unobservable in the aggregate. (Clapp, 1990)
Market value is observed only for the relatively small number of properties (folios) that change
hands in arm’s length transactions each year. Assessed values may fail to reflect accurate market
values in the short term but tend toward (correct to) market values over longer-term periods.
Research suggests that “[assessed value] and [repeat sales] methods are substantially similar over
131
a seven-year period. But [that] the repeat sales] method is inefficient because it uses a relatively
small subset of the data.” (Clapp & Giaccotto, 1992) The same study suggest that even rich repeat
sales dataset remain inefficient in predicting aggregate value.
Greater understanding of the nature and extent of variation in assessment
methodology and policy from one tax jurisdiction to another, and the degree to which aggregate
assessed value may vary from aggregate market value within any given geography would provide
additional nuance to these finding as well as having implication for practice. The extent to which
rising assessed values may lag rising market values, for example, may represent opportunity in the
nature of low hanging fruit with respect to value capture objectives.
Assessed valuation as a measure of value
Although assessed valuation is based on underlying market value, it is an imperfect,
and frequently lagged reflection/indication of current market value. The deficiencies of assessed
valuation as proxy for market value, particularly in real time, are mitigated by two considerations
in the context of this study.
This analysis is motivated in part by considerations related to the potential for value
capture techniques to provide an important source of funding for transit infrastructure projects.
Several of the mechanisms and strategies through which some portion of transit infrastructure-
induced value creation might be taxed for such purposes are dependent on assessed valuation
either directly or indirectly. Tax increment financing (TIF) is an example. In many applications
related to consideration of value creation and/or value capture, particularly from the perspective
of state or local government, assessed valuation may be the most appropriate measure.
A second mitigating factor is that this study considers neither absolute market nor
assessed values but differential rates of change in assessed value over time. The somewhat murky
disconnect between assessed values and coincident market values is only problematic for these
132
purposes if market and assessed valuations continue to diverge over time. Some researchers
conclude that such concerns regarding the reliability of [assessment] data on property valuations
in terms of variance between taxable and actual market value is effectively controlled through the
very large number of folios (properties) examined (Weinstein et al., 2002).
Ultimately, assessed value fulfills the requirements necessary for value-representative
statistical analysis: “six criteria which need to be satisfied from a statistical viewpoint: regular
availability, representativeness, homogenous comparability, unbroken and unchanging
description, length of series and data frequency” (Case & Wachter, 2005).
An observation regarding reassessment methodology
An additional factor contributing to the lumpiness of assessed valuation as a measure
of market value at any one moment in time is the spasmodic and non-uniform nature of value
reassessment. Approximately 4.5% of residential properties changed hands in 2015, the last year
of data availability in this study (Realtors, 2017). This means that when and where property
values are reassessed for tax purposes only when there are transactions, such reassessment affects
only a small proportion of total real property inventory. Complicating this fact is that in many
jurisdictions, such as those in California, conveyance of a partial interest in real estate (say, 50%)
results in reappraisal of only that portion of the property conveyed, leaving the appraised value of
the un-conveyed portion intact. Wholesale market-wide reassessment often occurs only every
two to five years, and can occur as infrequently as every ten years in some cases. Once again,
although appraised value will follow market value over long periods of time, aggregate assessed
value and real-time market value will be non-uniform at any moment in time.
Structural bias in data aggregation and reporting
CoreLogic has been accumulating and reporting assessed valuation data at the
individual parcel-/folio-level for 30-years. Due to limited market demand for historical data and
133
the cost of maintaining databases comprised of many millions of individual folio records,
CoreLogic chooses to maintain records only for the “current year” (2015 at the time this study
was undertaken in 2016) and the 10-years prior to the “current year” (2005 through 2014 in this
case). Variables capturing various aspects of assessed value records in prior years are associated
with “current year” folio numbers. Historical records are maintained and reported only for folio
numbers extant in the “current year” (2015). This data structure presents cause for concern both
when historical folios (parcels) are subdivided into smaller parcels (or units) with multiple
individual folio numbers and when multiple parcels are assembled into a single parcel within a
single folio number. In either case, if consistency of folio number designation is not maintained
by tax assessors over time, the available data structure may be biased toward higher aggregate
assessed value within discrete geographical boundaries over time. Analysis of sample data within
various jurisdiction suggests that this may not be a material problem, but it is one that must be
disclosed. In general, folio numbers appear to be very stable over time.
Missing data
A potential source of bias results from the fact that across the 21 transit systems under
consideration, specific stations were eliminated from analysis. These included 1) stations that
predominantly served only a single institutional or commercial use such as a government
complex, museum, campus, entertainment venue, or shopping mall; 2) stations that were
significantly multimodal (beyond LRT and bus service), including heavy rail or commuter rail
service, ferry service, etc.; and 3) where there were significant numbers of missing values under
primary dependent variables. Stations eliminated from analysis for one or more of these reasons
are identified in APPENDIX II
134
Measurement and reporting error
A potential source of bias within the assessed valuation data may be referred to
broadly as measurement and reporting error. Between treatment and control observations,
approximately 17-million individual observations of total assessed value and assessed value of
land and improvements were recorded. Among these there were clear and obvious examples of
erroneous values. CoreLogic collects and aggregates assessed valuation data from the many
individual tax jurisdictions and through various mechanical means. Erroneous values may have
been reported or recorded by the tax assessor, during data collection or consolidation, or
somewhere in the process of aggregating data in CoreLogic’s uniform data format. In many
jurisdiction-specific data series, the same value, for total assessed value, for example, was
reported under two or more alternative headings. Values aggregated at the station-and transit
line- level for indication of obvious outliers. Specific variables reflecting values for total assessed
value and assessed value of land and improvements were selected based on minimization of
outliers and missing data. In all cases, the same variables were employed in both treatment and
control. Outlying data values were not managed, however, at the level of the 51-million
individual data points.
Transit Areas of Influence, treatment, control, and buffer area definitions.
To minimize the extent of treatment confounding control, and to minimize the extent
to which particular folios (large land parcels) might overlap treatment and control areas,
treatment and control areas are separated by a ½-mile buffer (illustrated in Figures 8 and 9
above). In addition to the buffers surrounding treatment areas, a buffer of ½-mile from the
centerline of subject LRT lines is excluded from control areas. Rationale for elimination of folios
within the ½-mile buffer area surrounding stations is that (notwithstanding APTA definition of
primary catchment areas within TAIs), any treatment effect of proximity to transit stations
135
including the benefits of mobility and accessibility being capitalized into real estate and reflected
through market prices and assessed valuation, does not and would not end abruptly at the ½-mile
radius. One practical effect of this is that comparing rates of change between control areas (at
greater distance from stations) and those areas most impacted/benefited by proximity to stations
is likely to overstate any aggregate treatment effect. On the other hand, this study is likely to
underestimate aggregate differential value creation in that to the extent there is a positive (value-
creating) treatment effect, it would likely extend beyond the arbitrary ½-mile radius.
Rationale for elimination of folios within the ½-mile buffer along rail line corridors
outside of TAIs is that proximity to LRT lines and rail traffic may have a negative effect on the
value of property not proximate to stations (within treatment areas). Folios within these buffer
areas are eliminated to avoid overstatement of any prospectively positive treatment affect.
External validity
This analysis addresses external validity (limited to that within the continental United
States) by studying and comparing differential value creation trends along 21 transit lines within
14 transit systems geographically dispersed across the United States and located within
significantly heterogeneous metropolitan markets. Transit line selection is limited, however, by
data availability in the 11-year period of 2005 through 2015, and those systems with new service
development within a timeframe relevant for study given available assessed valuation data. The
period under consideration reflects market conditions and other constraints and impacts specific
to that time period. Beyond this general concern is the fact that the period of analysis spans the
Great Recession and the many, perhaps confounding, impacts of the financial crisis and its
aftermath. There is reason to believe that markets were working inefficiently in some cases and
not at all in others during this period.
136
A longitudinal analysis extending many more years, both prior and subsequent to
treatment, would be more robust. An earlier (prospective) design of this study would have
included 30-years of observations (and more observable station areas as a result). Data prior to
2005 were simply not available, however, limiting both the robustness of the longitudinal
difference-in-differences comparison and the number of station areas under consideration.
Serial correlation
Another concern related to the limited longitudinal data series, and the fact that we
cannot observe pre-treatment values either within treatment or control areas for all cases, is that
of serial correlation of outcomes and resulting inconsistency of standard errors. Stations along
many lines are very close to each other (TAIs overlap in many cases). A result of which is that
value creation in one is likely to affect that in proximate stations in some cases (Hubert,
Golledge, & Costanzo, 1981). A 2002 paper by Bertrand, et. al., found that:
Most papers that employ Differences-in-Differences estimation (DD) use many years of data and focus on serially correlated outcomes but ignore that the resulting standard errors are inconsistent. … These conventional DD standard errors severely understate the standard deviation of the estimators. … Two corrections based on asymptotic approximation of the variance-covariance matrix work well for moderate numbers of states and one correction that collapses the time series information into a “pre”- and “post”-period and explicitly takes into account the effective sample size works well even for small numbers of states.
Unfortunately, we do not have a very long longitudinal series and cannot observe
“pre”- and “post”-periods in all cases.
Selection bias
This is a quasi-experimental study in the sense that repeated observations are made over
an 11-year period in both treatment and control groups (areas); treatment being comprised of
proximate LRT service delivered at/through newly-developed stations in specific locations. The
difference-in-differences analysis is a comparison between treatment areas (TAIs within ½ mile
of station centroid locations) and control areas (between 1-mile and 2-miles from station
137
centroids) prospectively less-impacted by new transit accessibility and utility. Definition of
treatment areas as those within a ½-mile radius of station centroids is motivated by a desire for
consistency with literature defining TOD areas of greatest impact, dimensions employed in
numerous other studies, and the Primary Catchment Area defined by APTA (APTA, 2009).
A truly experimental research design requires random assignment of treatment. A concern
with this quasi-experimental design is that assignment to treatment (e.g. design, entitlement,
funding, and development of the transit corridors and location-specific stations) is entirely
That the model only explains approximately 28% of observed variation in differential
value creation rates from station to station suggests that there are significant factors at play in
addition to the covariates considered herein. Some of these may include very fine grain sub-
market, regulatory, environmental, institutional, geographic, topographic factors as well as
153
neighborhood characteristics. Identification and understanding of such factors would be useful
for economic impact assessment and financial modelling related to consideration of new transit
(or other transportation) infrastructure investment.
Understanding predevelopment land speculation and timing of value creation
The original proposal for this research, based on the prospect of 30-years of assessed
value data and 1,100 relevant U.S. light rail stations as opposed to 11-years of data and the 229
stations observed in this study, included measuring differential rates of differential value creation
prior and subsequent to the putative “treatment” defined as commencement of service at each
station. Reduction in scope of available assessed value data meant that more than a few of the
commencement of service dates for individual stations and lines fell outside the date range of
available data. This analysis was dropped in an effort to maintain as broad a scope (and as many
stations under observation) as possible.
The opportunity now exists to consider that subset of stations with commencement
of service dates somewhere between 2005 and 2015 in an effort to understand and model the
extent and timing of anticipatory predevelopment land speculation and/or other value creation.
Implications for practice
Public-sector investment in transportation infrastructure has long been a driver of
private-sector investment and value creation. Ongoing consideration of prospectively massive
transportation infrastructure investment and re-investment within the United States combined
with increasing interest in value capture as a finance strategy focus ever greater attention on
inducement of value creation through new LRT (and other) infrastructure investment. Any
seductive presumption that “if we [the public-sector] build it [new transit infrastructure], they
[producers of incremental value creation] will come,” is inadequate and potentially dangerous.
154
Economic, market, and station-specific sub-market forces appear to wield significantly greater
determinative power with respect to value creation than does infrastructure investment itself.
Development that occurs surrounding newly-developed LRT stations is often more
intensive, more complicated, and perhaps more risk-laden than competitive (non-transit-
influenced) projects. Policy makers, planners, transit professionals, and financiers may need to
facilitate earlier and more comprehensive strategic planning and engagement in the value
creation process in order to realize the significant potential value creation that can occur under
the right circumstances. Fruitful public-private partnership and strategic engagement will require
much greater understanding of the economic, market, and sub-market forces that determine the
extent of differential value creation than may currently exist. Value capture strategies that
encourage and reinforce continued value creation, rather than creating competitive disadvantage
and economic disincentive, will remain entirely dependent on robust value creation and,
therefore, on successful value creation strategies.
155
APPENDIX I – Survey Instrument
156
157
158
159
160
161
APPENDIX II – Stations Excluded from Analysis
The following stations were eliminated from analysis due to insufficient treatment
area data, insufficient control area data, extensive overlap of folios within treatment area with
those in adjoining station areas, missing data, insufficient control area comparables, and/or
location of station on peninsula surrounded by water bodies.
§ Dallas o North Carollton/Frankfort o Trinity Mills
§ Denver o Pepsi Center o Sports Authority Field o Theater District o Colfax at Auraria o Auraria West o 10th & Osage o Alameda o Louisiana & Pearl o I25 & Broadway
§ Houston o Memorial Hermann
Hospital § New Jersey Transit – Hudson-
Bergen Line o 22nd Street o 8th Street
§ New Jersey Transit – River Line o Aquarium
§ Los Angeles o La Brea o La Cienega
§ Minneapolis o 28th Avenue o American Boulevard o Bloomington Central o Fort Snelling
§ Phoenix o Center Parkway o Priest Drive & Papago
Park § Portland
o Delta Park/Vanport o Mount Hood Avenue o NW 5th/6th & Couch/Davis o Pioneer Courthouse o PSU South, SW 5th/6th o SW 5th/6th & Jefferson o SW 5th/6th & Oak/Pine o SW 6th & College o SW 6th & Montomgery
§ Salt Lake City o 5600 West Old Bingham
Highway o Decker Lake o Fort Douglas o South Jordon Parkway o University Medical Center o University South Campus
§ Seattle o SODO
162
APPENDIX III – GIS Data Processing
The following description of GIS data processing and methodology is provided by
Phillip McDaniel, GIS Librarian, Davis Library at the University of North Carolina at Chapel
Hill.
TRANSIT STATION TERRITORIES
Background
One of the main goals of this effort was to assign folios to transit stops based on
proximity. This assignment of folios was done at two different geographic specifications:
within ½ mile, and between 1 and 2 miles. The transit stations were processed to create two
territories per station: a ½ mile Euclidean buffer (hereafter referred to as TA)(maximum total
area = 0.785 sq miles), and a 1 to 2 mile ring buffer (hereafter referred to as CA)(maximum
total area = 9.426 sq miles).
The requirements were different for the construction of the TAs and CAs,
necessitating different processing for each type. Specifically, the TA for each transit stop was
constructed such that there would be no overlap with the TAs of adjacent transit stops (see
diagram, below). In those areas where there was overlap between TAs, the overlapping area
was split (roughly evenly), with a portion being assigned to each of the overlapping TAs. The
partitioning of space was done in this way so that CoreLogic folios would be assigned to one,
and only one, transit stop- that is, there is no double, triple, etc, counting of folios within the
TAs. For example, also in the diagram that follows, even though it falls within ½ mile of both
transit stop A and B, the blue dot is assigned to stop A since it it closer to stop A than it is to
stop B (likewise, the orange dot is assigned to stop B due to its closer proximity to stop B).
163
The CAs were not bound by this overlap restriction, however, so the number of
overlapping CAs would be higher or lower depending on the density of transit stops (i.e. cities
with a large number of transit stops in close proximity could have areas that are overlapped by
many CAs). Further, any folios that fell within an area of CA overlap were assigned to each of
the overlapping CAs. For a conceptual example, in the diagram above (even though it doesn’t
show ring buffers), since the blue point is within an area of overlap between both A and B, it
would be assigned to both transit stops (likewise for the orange point). In the more densely
populated urban study sites, some folios might be assigned to 10 or more transit stops.
Processing: Control Areas (CAs)
The transit stop points for all cities were first projected from a geographic
coordinate system (GCS) to a projected coordinate system (Albers Equal Area Conic, USGS
version) so that distances and areas could be accurately calculated. Once projected, the CA
buffers were created using the [BUFFER] tool in ArcMap 10.4.1. This tool generates
Euclidean (i.e. as the crow flies) buffers at user defined intervals. Since the CA buffers were
rings that only covered areas between 1 and 2 miles from the transit stops, a two step process
was employed:
1. Create 1 and 2 mile buffers around each transit stop (A, below).
2. Erase the 1 mile buffer from the 2 mile buffer (B, below). This step clips out the 1 mile
164
buffer from the middle of the 2 mile buffer, resulting in a 1 to 2 mile ring around each
station.
A final processing sequence was performed on the CAs in order to remove the
areas in close proximity (½ mile) to transit lines:
1. Using the [BUFFER] tool, creat ½ mile buffers constructed around transit lines (shown in
the dashed line, below).
2. Erase the transit line buffers from the 1 to 2 mile CA buffers. This step clips out the ½
mile transit buffers from the 1 to 2 mile CA buffers (see diagram, below). This process of
clipping out the ½ mile transit line buffers effectively removes territory from the CAs, so
any folios that fall in those areas will not be included in the CAs for transit stops.
Processing: Treatment Areas (TAs)
165
In addition to the processing described above, the transit stops were processed in
order to create non-overlapping TAs. Instead of simply creating ½ mile buffers around each
station, the TAs were further processed to eliminate any areas of overlap with adjacent transit
station TAs. Specifically, Thiessen polygons were constructed, then a UNION function was run
with the ½ mile buffers to create non-overlapping territories for each transit stop. Thiessen
polygons are polygons whose boundaries define the area that is closest to each point relative to
all other points:
This Thiessen polygon and UNION approach was used for three main reasons:
1. Due to processing complexities involved with precisely bisecting the areas of overlap
between the ½ mile transit stop buffers
2. Time intensive to manually interpret and delineate boundaries
3. Difficult to accurately reproduce results or extend research to other study sites without a
clearly defined workflow
The diagram below provides a conceptual overview. The ½ mile transit buffers are
defined by the dashed lines surrounding the circles. The Thiessen polygons are defined by the
thin gray lines. When a UNION is performed on these two layers, the Thiessen polygons and ½
mile buffers create a new set of polygons that contain all overlap combinations (i.e. a Venn
166
Diagram). For each transit stop, the TA is then determined by areas that are both within its
Thiessen area and within its ½ mile buffer.
In a few instances, due to the configuration of transit stops, there are slight
differences between the bisection of overlapping ½ mile buffers and Thiessen polygon
territories. In the above diagram, these discrepancies are indicated by the arrows to the right.
In this scenario, for the two easternmost stops, there are some areas within the ½ mile buffer of
the red transit stop that are actually closer to the green transit stop, and vice versa. When this
occurs, the default is to assign those areas based on the transit stop Thiessen polygon that they
fall within.
The processing of the TAs was as follows:
1. Create ½ mile buffers around each transit stop using the [BUFFER] tool in ArcMap.
2. Create Thiessen polygons around each transit stop using the [CREATE THIESSEN
POLYGONS] tool in ArcMap. The resulting polygons contain all of the attributes
from the input point features (e.g. station name, station ID, etc.)
3. Combine the 1.2 mile transit stop buffers and the transit stop Thiessen polygons using
the [UNION] tool in ArcMap. The resulting polygon file contains features for all
possible overlaps, listed below:
a. ½ mile buffer - Thiessen
167
b. No ½ mile buffer - Thiessen
4. Each polygon overlap feature also contains the attributes of the input features. For
example, in scenario a above (½ mile buffer - Thiessen), a feature would have
information from both the ½ buffer input and the Thiessen polygon input. In instances
where the ½ mile buffer for transit stop ABC is fully within the Thiessen polygon for
ABC, there will only be one record in the UNION file with ½ mile buffer attributes for
transit stop ABC. However, if transit stop ABC’s ½ mile buffer were to overlap with
transit stop DEF’s ½ mile buffer, there will be two records in the UNION file for transit
stop ABC- one for the intersection with its own Thiessen polygon, and one for its
intersection with the Thiessen polygon of DEF.
In the example table below, several overlap scenarios are shown:
1. ABC intersects DEF
2. DEF intersects ABC and GHI
3. GHI intersects DEF
4. JKL does not intersect any other ½ mile buffers
½ Mile ID Thiessen ID
ABC ABC
ABC DEF
DEF DEF
DEF ABC
DEF GHI
GHI GHI
GHI DEF
JKL JKL
168
In the above scenario, the TAs would then be constructed based on a shared Thiessen
ID. The color coding shows the UNIONed polygons that will be combined to form the TA for
each transit stop:
1. Orange is the TA for ABC
2. Green is the TA for DEF
3. Blue is the TA for GHI
4. Purple is the TA for JKL
CORELOGIC FOLIO ASSIGNMENT
Compared to workflow to construct the CAs and TAs, the processing to associate
folios with station CAs and TAs was relatively straightforward (I will lean on Matt to write
some sentences about the specific heavy lifting involved to pull meaningful information from
the massive data tables). As stated above, the main difference between the TAs and CAs is
that an individual folio could only be assigned to a single TA, whereas an individual folio
could be assigned to as many TAs as it falls within.
The first step in processing the folios was to plot the folios based on their X and Y
(longitude and latitude) coordinates in order to create a point GIS file. Plotting the points was
a simple process, requiring only that the X and Y fields be properly specified so that the points
could be displayed in their correct geographic locations. This point file was subsequently used
as an input to a second tool- [SPATIAL JOIN].
The tool used to join the folio data to the transit station TAs and CAs was [SPATIAL
JOIN]. This tool joins attributes from one feature to another based on the spatial relationship. In
this instance, the transit stop TAs and CAs were spatially joined to the folios.
169
For the TAs, the join type was one-to-one, where a folio could only fall within a
single station’s TA. The output was a shapefile (a common vector GIS format) containing
points for the folios that intersected a TA, with the transit station ID that was intersected
appended to the folio attributes. In this file, which contained only the folios that were within a
TA, each folio was only represented once.
For the CAs, the join type was one-to-many, where a folio could fall within any
number of CAs. The output for this was a shapefile containing points for the folios intersected a
CA, with the transit station ID that was intersected appended to the folio attributes. In this file,
however, each folio could be included multiple times, once for each intersection with a different
CA. For each instance of a folio, a different station’s CA was appended to the folio attributes.
Much of what Matt did involved using the TA and CA folio files to summarizing the
folios by transit station ID. For example, for each transit ID in the TA folio file, calculate the
mean property value. Or, for each transit ID in the CA folio file, calculate the mean value per
acre.
170
APPENDIX IV - Selection of demographic covariates
The original model specification included a greater number of independent variables than
could be accommodated due to absolute constraints of the ANOVA statistical analysis software. A
large number of prospectively relevant demographic covariates were obtained from the 2010
American Community Survey data at the census block-level. These were analyzed in preliminary
models and wherein total assessed value was regressed on them to identify and eliminate variables
with significant collinearity. The resulting list of demographic (control) covariates included in the
final model comprised median household income, gross rent burden greater than 30%, persons per
household, percent of dwellings owner occupied, total vehicles per household, and dwellings
percent vacant.
§ Age-Sex: SEX BY AGE
§ Age-Sex: MEDIAN AGE BY SEX
§ Educational Attainment: SEX BY EDUCATIONAL ATTAINMENT FOR THE
POPULATION 25 YEARS AND OVER
§ Income: MEDIAN HOUSEHOLD INCOME IN THE PAST 12 MONTHS (IN 2015
INFLATION-ADJUSTED DOLLARS)
§ Income: PER CAPITA INCOME IN THE PAST 12 MONTHS (IN 2015 INFLATION-
ADJUSTED DOLLARS)
§ Employment Status: EMPLOYMENT STATUS FOR THE POPULATION 16 YEARS
AND OVER
§ Housing: OCCUPANCY STATUS
§ Housing: TENURE
§ Housing: TOTAL POPULATION IN OCCUPIED HOUSING UNITS BY TENURE
171
§ Housing: AVERAGE HOUSEHOLD SIZE OF OCCUPIED HOUSING UNITS BY
TENURE
§ Housing: TENURE BY VEHICLES AVAILABLE
§ Housing: AGGREGATE NUMBER OF VEHICLES AVAILABLE BY TENURE
§ Housing: GROSS RENT AS A PERCENTAGE OF HOUSEHOLD INCOME IN THE
PAST 12 MONTHS
Of these, demographic covariates included in the final model included Median
Occupied, Total Vehicles Per Household, and Dwellings Percent Vacant.
Selection of independent covariates for inclusion in model(s) Covariate evaluation began with identification of potentially confounding variables (which
variables resulted, initially, in a highly complex ANOVA model). Although inclusion of all
available covariates allowed examination of each possible covariate’s effect, it also interfered with
model execution. This was exemplified by the model’s inability to calculate many of the two- and
three-way interactions initially considered. Additionally, between-effects for several single
variables could not be estimated. These included the transit agency goals and objectives, and value
creation and value capture strategies. Following removal of variables for which between-effects
could not be estimated, the model was reconstructed excluding three-way (or greater) interactions
excluded. The resulting, somewhat reduced, model was assessed to evaluate remaining variables
for collinearity. Vehicles per renter household and vehicles per owner-occupied household were
too highly correlated to be included together in the model. Vehicles per owner-occupied household
had the stronger effect and was retained. Vehicles per renter household excluded. Other variables
pre-assumed to be correlated were not problematic upon examination. These included parking
spaces and station character, as well as median household income and per capita income. In these
172
cases, all variables were retained. The final four models controlled for station-specific
characteristics and transit agency goals and objectives as well as transit system (location) and
demographic characteristics included 17 covariates, in addition to the effect of time and treatment.
173
APPENDIX V – Transit Agency Survey Data – Descriptive Statistics
Statistics descriptive of survey responses are reflected in Tables 13 - 16.
Table 22: Descriptive statistics – station characteristics
n %
Station Characteristics
Station Design Elevated 25 10.9 At Grade 191 83.4 Open Cut 9 3.9 Underground 4 1.7
Station Typology Downtown-CBD 34 14.8
Urban Center 17 7.4 Urban Neighborhood 89 28.9 Suburban Town Center 18 7.9 Suburban Neighborhood 46 20.1
Campus, Entertainment, Special 21 9.2 Other 4 1.7
Station Character Walk-and-Ride 146 63.8
Park-and-Ride 83 36.2
174
Table 13: Descriptive statistics – transit line design/development objectives n %
Transit line design/development objectives
Serving existing commuters based on pre-transit residential and employment patterns
No 80 34.9 Yes 149 65.1 Serving anticipated residents and workers drawn to new development near transit stations
No 119 52.0 Yes 110 48.0
Inducing transit ridership No 55 24.0
Yes 174 76.0 Relieving roadway congestion
No 127 55.5 Yes 102 44.5
Economic development
No 84 36.7 Yes 145 63.3
Growth management, environmental, or other public policy mandates or objectives No 160 69.9
Yes 69 30.1 Political or public initiatives or agenda
No 117 51.1 Yes 112 48.9
Private sector commercial interests or development initiatives No 227 99.1
Yes 2 0.9
175
Table 14: Descriptive statistics - value creation strategies n %
Value Creation Strategies
None: no specific value creation strategies were implemented
No 176 76.9 Yes 53 23.1
Strategic partnership between public and private interests No 199 86.9
Yes 30 13.1 Land use, zoning, entitlement enticements
No 142 62.0 Yes 87 38.0
Complementary (non-transit) public infrastructure investment No 174 76.0
Yes 55 24.0 Land assemblage, acquisition, or other real estate strategies?
No 160 69.9 Yes 69 30.1
Other No 218 95.2
Yes 11 4.8
176
Table 15: Descriptive statistics - value capture strategies n %
Value Capture Strategies
Joint Development No 201 87.8
Yes 28 12.2 Negotiated Exactions No 224 97.8
Yes 5 2.2 Tax Increment Financing (TIF) No 207 90.4
Yes 22 9.6 Special Assessments Impact Fees No 207 90.4
Yes 22 9.6 Land Value Taxation No 0 0.0 Yes 0 0.0 Naming Rights No 224 97.8
Yes 5 2.2 None: No Value Capture
No 63 27.5 Yes 166 72.5
177
Table 16: Descriptive statistics - transit agency perspectives n %
Transit Agency Perspectives
How significant/important was value capture strategy to financial viability or project success?
Not Important 187 81.7 Slightly Important 18 7.9 Moderately Important 6 2.6 Significantly Important 9 3.9 Very Important 9 3.9
How important are public-private value creation strategies to the success of transit-influenced or transit-oriented development?
Not Important 30 13.1 Slightly Important 22 9.6 Moderately Important 34 14.8 Significantly Important 112 48.9 Very Important 31 13.5
How important are public-private value creation strategies to the success of transit infrastructure investments and projects?
Not Important 31 13.5 Slightly Important 22 9.6 Moderately Important 64 27.9 Significantly Important 77 33.6 Very Important 35 15.3
How important is value capture to the success of transit infrastructure investments and projects?
Not Important 51 22.3 Slightly Important 12 5.2 Moderately Important 68 29.7 Significantly Important 85 37.1 Very Important 13 5.7